DALL-E is one of the innovative generative AI platforms blurring the lines between human- and computer-generated creativity. Here’s an overview of DALL-E, how to use it, and what you should know to make it work for you.

Table of contents

• What is DALL-E?
• Who created DALL-E?
• Evolution of DALL-E
• How DALL-E works
• Is DALL-E free?
• How to use DALL-E
• Use cases and applications
• Benefits of DALL-E
• Shortcomings of DALL-E
• Conclusion

What is DALL-E?

DALL-E is a generative AI platform that turns text prompts into images. DALL-E can process natural language, so you don’t need any special coding or image-editing abilities to use it. You can enter prompts that describe your desired image’s subject, style, framing, and other characteristics, and DALL-E will produce a visual representation that matches your description. It can also edit existing images.

The name DALL-E was inspired by a combination of the names of two well-known figures: the Spanish surrealist artist Salvador Dalí and WALL-E, the robot in the 2008 Pixar movie of the same name.

Who created DALL-E?

OpenAI, the same company behind ChatGPT, created DALL-E. OpenAI is an AI research company founded in 2015.

Open AI released DALL-E in January 2021. It released DALL-E 2 in September 2022 and DALL-E 3 in October 2023.

How has DALL-E evolved?

OpenAI announced its first image generation tool in 2020, and DALL-E has evolved from there. OpenAI’s first foray into image generation was called Image GPT. Image GPT provided the first proof that the GPT model could create images.

Then came DALL-E. The first iteration of DALL-E was based on a version of GPT-3—the large language model (LLM) that OpenAI released in 2020—adapted for image generation.

DALL-E creates believable images and accomplishes several tasks, some of which include:

• Modifying several characteristics of an object, such as the color and texture of a sphere
• Understanding framing, such as close-ups and wide angles
• Creating images of the same object from multiple angles
• Understanding geographic information and periods in history

What is DALL-E 2?

The next version, DALL-E 2, generates images with four times higher resolution than images generated by DALL-E. It handles composition and object placement more effectively, making elements like shadows and lighting appear more realistic. DALL-E 2 also introduced two new features for modifying existing images: inpainting and outpainting.

• Inpainting is when you erase a portion of an image and use AI to fill in the empty space with something else. For instance, you can remove a building from the background of a photo and replace it with a tree.
• Outpainting is when you expand the borders of an image with AI. For example, if you have a close-up image of your dog in a park and want to expand it to show the city skyline in the distance, DALL-E 2 does that with outpainting.

What is DALL-E 3?

DALL-E 3 is a significant improvement over its predecessor in several ways. For starters, it’s better at interpreting prompts. Previous versions would skip over words and descriptions. You had to become good at prompt engineering to get the image you wanted. DALL-E 3 understands nuance and context better and can follow more complex prompts. Its responses are more accurate, and its images are more coherent. Ultimately, its output better aligns with what people want.

DALL-E 3 also includes more sophisticated security measures. For example, it prevents explicit, aggressive, or discriminatory images. To prevent people from creating images that infringe on copyrights and violate intellectual property, DALL-E 3 doesn’t generate images that resemble living public figures or that mimic the style of popular artists and brands. DALL-E 3 also allows creators to opt out of having their images used for training future models.

Inclusion with existing AI tools

DALL-E 3 is included natively with ChatGPT and Microsoft Image Creator from Designer (formerly Bing Image Generator).

This means that if you have a premium ChatGPT subscription, you can generate images as part of your conversation with the chatbot. With this capability, you don’t just have to write straightforward prompts. You can ask questions or give directions, and ChatGPT can hand them to DALL-E to generate an image.

For example, you might say, “I just moved to Arizona, and everyone keeps talking about something called a haboob. What does that look like?” ChatGPT can process your question and generate a prompt for DALL-E. DALL-E will then create images of a haboob, which is a dust storm that occurs in dry areas like Arizona.

ChatGPT will also elaborate on your prompts to provide DALL-E with more detail. If you write a prompt that says “Create an image of two cats sitting on a chair, in a vintage photographic style,” ChatGPT may refine your prompt to this: “Create a black-and-white vintage photograph of two cats sitting on a green sofa chair. One cat is a tabby, and the other is gray all over. The two cats are sitting side by side.”

How DALL-E works

At a basic level, DALL-E uses deep learning to understand the relationships between images and text, allowing the model to output new images for a text prompt. The specific generative AI models behind DALL-E are constantly evolving.

DALL-E 1

DALL-E 1 (also called DALL-E) uses a version of GPT-3, OpenAI’s LLM, that was trained to generate images from text descriptions. This model is based on a transformer architecture. Just as ChatGPT generates text by predicting each word one by one, the original version of DALL-E generates images by predicting each pixel.

DALL-E 1 generates many candidate outputs for a single prompt. A second AI system, called CLIP (Contrastive Language-Image Pretraining), is used to select the best one. CLIP, just like DALL-E 1, is trained on a large image and caption dataset. However, the goal of CLIP is to understand how closely a given image and text caption are related.

DALL-E 2

DALL-E 2 generates images using a diffusion model rather than an LLM for improved image quality and accuracy.

This approach trains a model to take noisy images, where pixels have been distorted in a random way, and incrementally remove the noise to reveal a clear image. Then you can give a model a set of pixels plus noise—which represents some underlying image features, such as “a cat in a top hat”—and the model will construct a new image from scratch.

DALL-E 2 uses CLIP to understand the text in a user’s prompt and map it to image features. This information is passed to the diffusion model, allowing it to generate an output that fits the user’s prompt.

DALL-E 3

Little is known about the architectural differences between DALL-E 2 and DALL-E 3. This is because OpenAI has not shared this information publicly. However, DALL-E 3 almost certainly uses a diffusion model, as this is widely accepted as the state-of-the-art technique for image generation.

There is speculation that DALL-E 3 uses more advanced diffusion techniques and may be using an LLM (rather than a smaller model like CLIP) to understand relationships between images and text.

Is DALL-E free to use?

DALL-E is available with a paid ChatGPT subscription, which is offered in several tiers for individuals and businesses.

You can access DALL-E for free with Microsoft Image Creator from Designer (formerly Bing Image Generator). Image Creator is also available through Copilot, which is Microsoft’s chatbot.

Tips for using DALL-E

Here are some tips for getting the best results with DALL-E:

Be descriptive

The more precise your prompt, the better DALL-E’s output will be.

• Provide a clear description of the main subject; for example, “a blue microfiber couch” instead of just “a couch.”
• Explain the setting, such as “on a tropical beach,” “in a 1970s house,” or “inside an elementary school gym.”
• Detail any action, like “the sun is setting,” “a dog is napping,” or “a kite is flying.”
• Describe the image format, such as “photorealistic,” “painting,” or “pencil sketch.”
• Tell DALL-E which style you want; for example, “black and white,” “abstract,” or “art deco.”
• Include camera angle and focal distance, like “aerial view,” “close-up,” or “wide-angle.”
• Provide lighting details, such as “deep shadows,” “flash,” or “backlit.”
• Describe the mood; for example, “romantic,” “gritty,” or “dreamy.”

Be experimental

There’s no textbook or perfect way to use DALL-E. The best way to get the results you want is to take an experimental approach to using it.

• Make minor tweaks to your prompts to see if you get better results. Try using variations of the same words to see if it alters your results.
• Find the right balance of details. If your prompts are too detailed, DALL-E may not know which ones are most important. Play around with the complexity of your prompts to find your sweet spot.
• Brace for mistakes and failures. DALL-E can get offtrack. Take each failed response as a learning opportunity. Finding out what doesn’t work is just as important as finding out what does.

DALL-E use cases and applications

People use DALL-E for many applications in business and personal settings.

Marketing and business communications

• Creating images for blogs, social media posts, and websites
• Designing advertisements, such as fliers and posters
• Designing logos and brand elements
• Creating one-of-a-kind stock photos
• Designing product packaging

Conceptualization

• Designing physical products
• Rendering architectural models
• Ideating other creative projects, such as animation, storyboards, and interior design
• Testing out creative ideas in different styles

Educational content

• Creating visual aids like infographics and diagrams
• Depicting historical events
• Visualizing scientific processes that you can’t see with the naked eye, such as chemical reactions
• Creating images tailored to an individual student’s specific needs, interests, or learning style

Art and design

• Creating custom artwork for your home or party decor
• Designing cover art for books, albums, or movies
• Creating art to sell on products like T-shirts, bookmarks, and prints
• Creating reference images to use as inspiration for other art mediums, like fashion design
• Designing elements, such as background textures, to incorporate into other forms of artwork

Modifying existing images

• Adding more subjects to an image
• Adjusting the background
• Changing the aspect ratio
• Emphasizing certain objects
• Removing an object and replacing it with something else

Benefits of using DALL-E

DALL-E offers numerous advantages, including the ability to choose from multiple responses, use the platform alongside other AI tools, and remove barriers to art and design.

Generates multiple images per prompt

DALL-E generates four images per prompt, so you can choose the one that best suits your preferences. It modifies the prompt slightly for each image and expands on it to add more detail.

For example, if you enter a generic prompt like “A comic-book-style image of a dark alley,” DALL-E will rephrase your prompt and add details like the style of buildings in the scene, the framing of the image, or the predominant colors. You can see DALL-E’s prompt variations by clicking on each image.

Integrates with ChatGPT and Microsoft Copilot

You can access DALL-E through chatbots that you may already be using. It’s convenient to generate text and images all inside of one tool. Also, since these are chatbots, the images you generate can be part of a longer conversation.

For example, suppose you’ve been using ChatGPT to create an agenda for a baby shower. In that case, you can also use DALL-E to make the images for the invitations. Since it’s all part of one conversation, ChatGPT can incorporate some of the details of your agenda into the invite.

Makes design more accessible

Design software and photography equipment can be expensive and challenging to learn. DALL-E makes image generation more accessible for the average person.

• A small business owner can create custom brand assets, like photos and product images that would have previously been out of reach.
• Hobbyists in areas like woodworking and sculpting can draft visualizations of their concepts without investing in costly software.
• People and organizations from underrepresented groups or with niche hobbies can create images that speak to their interests.

Shortcomings of DALL-E

Despite its capabilities, DALL-E does have some limitations.

Unpredictability

Since DALL-E generates every image from scratch, it can be unpredictable. Suppose you have specific requirements for object placement or brand standards. In that case, DALL-E may not always incorporate those standards in its results.

Also, slightly adjusting your prompt may result in a significantly different output. This is especially challenging when changing an image DALL-E has already created.

Biases

All generative AI deals with biases, and DALL-E is no different. DALL-E is subject to generating responses that reflect biases about race, gender, class, and even certain languages or countries. DALL-E was trained primarily on data from the US, so it often reflects American culture, values, and biases.

Using certain adjectives may lead to stereotypical results. For instance, if the prompt contains words like emotional or sensitive, the output may be associated with a woman. At the same time, words like tough or intellectual may lead to results that feature men.

Cost

DALL-E comes at a cost unless you use Microsoft Image Creator, which may be inconvenient, depending on your preferences.

If you prefer using ChatGPT over Microsoft’s AI platforms, you’ll have to pay to access DALL-E.

What’s next for DALL-E and AI image generation?

You can use DALL-E to fuel creative brainstorming, streamline design processes, or simply have fun. It’s one of the many generative AI platforms that allows you to create in new ways. Because it’s integrated with existing AI platforms like ChatGPT and Microsoft Image Creator, you can create images and generate text all within a single tool.

When using DALL-E, it’s important to be mindful that all generative AI is prone to producing biased responses. Knowing the limitations of DALL-E allows you to find the best ways to use it and get the images you want.

New capabilities, features, and competitors are constantly emerging. Anyone who wants to use generative AI—whether for business, personal, or educational purposes—should keep tabs on the latest developments. We’ll keep covering the significant changes in generative AI, so keep up with the Grammarly blog to stay in the loop.

The post DALL-E 101: What It Is and How It Works appeared first on Grammarly Blog.

As Google’s answer to ChatGPT, Gemini can change how you search the internet and interact with Google services and apps. Learn what Gemini is, how to use it, and which potential shortcomings to avoid.

Table of contents

• What is Gemini?
• How Gemini works
• Gemini release date
• Is Gemini free?
• How to use Gemini
• Advantages of Gemini
• Disadvantages of Gemini
• Conclusion

What is Gemini?

Google Gemini, previously known as Google Bard, is an AI-powered chatbot. It uses machine learning and natural language processing to provide humanlike responses to text, image, and audio prompts.

Gemini performs several functions. You can ask it questions or make requests, and it will respond with text, code, or images. Gemini integrates with Google apps and services, utilizing the vast database of Google’s search engine to inform its responses.

How does Gemini work?

Gemini relies on a subset of machine learning called a large language model (LLM). LLMs are capable of efficiently ingesting and parsing through large volumes of data. Here’s an overview of how Google’s LLM innovations led to the development of Gemini.

What makes AI models tick

First, let’s look at how generative AI works more broadly. Data scientists and researchers start by training a model on vast amounts of data. By mapping the relationships among words, phrases, and images in the training data, the model can make predictions about what prompts mean and which response it should generate. Each word in a sentence or pixel of an image is a prediction.

To ensure the responses meet users’ needs, generative AI models typically undergo a fine-tuning stage during which they are given additional, specific data (such as a database of conversations) and human feedback.

Large Language Models, including those that power Gemini and ChatGPT, use a specific type of model architecture called a transformer. Google researchers introduced the transformer architecture in 2017, and it became a game changer in machine learning for several reasons:

• It requires fewer computational resources.
• It models the relationships between words in a sentence, regardless of the word order, to assign context and meaning.
• It processes multiple words at the same time, accelerating the training process.
• It supports multiple types of inputs and outputs, including text, images, and audio.

Google models used to power Gemini

Google has used several LLMs to power Gemini.

Gemini was initially based on Google’s Language Model for Dialog Applications (LaMDA):

• Announced in 2021
• Trained on publicly available dialogue and web content
• Fine-tuned by humans, who rated responses for sensibleness, specificity, and interestingness

Google replaced the LaMDA model with the Pathways Language Model (PaLM 2):

• Trained in 100 languages
• Enabled Gemini to generate and debug code
• Used a more extensive training dataset, including books, conversational data, and mathematical content

In December 2023, Gemini (then known as Bard) was moved to the Gemini LLM:

• Trained with multimodal data (text, images, and audio)
• Can understand more context and nuance since data is coming from more than text-only sources
• Can analyze large amounts of complex information, such as an annual financial report

When was Google Gemini released?

Gemini was released in March 2023 in what Google called “an experimental phase.” The official public release was limited to the US and UK; you had to sign up for a waitlist.

The international release was announced in May 2023. Gemini is now available in 40 languages and 230 countries.

Is Google Gemini free to use?

Google offers free and paid versions of Gemini. You can access Gemini via the web application or iOS and Android apps.

The free version offers all of the basic features:

• Text-based prompts and generation
• Ability to upload and generate images
• Ability to search Google apps and services

The paid version, Gemini Advanced, offers more powerful features:

• Advanced version of the AI model, which is designed for more complex tasks
• Ability to have longer conversations
• Ability to use Gemini inside Google apps like Gmail and Docs
• 2TB of storage

How to use Google Gemini

The sophistication of Gemini’s AI models and the breadth of Google’s existing services enable you to use it in many ways.

Text generation

Enter a prompt, and Gemini will respond with conversational text. You can generate text for various business, personal, academic, or creative applications.

Examples of text generation tasks include:

• Drafting content for emails, letters, and other forms of correspondence
• Creating educational content, such as speeches, study guides, presentations, and lesson plans
• Translating text from one language to another
• Drafting business communications like proposals, website content, and memos
• Providing tips to revise or improve existing written content
• Writing creative content, such as social media posts, storylines for games, and prompts for journaling exercises

Gemini is just one of many AI-powered text generation tools. Alternative platforms also allow you to generate text inside other apps. Grammarly, for example, can help you write text inside apps like Microsoft Word or Gmail, so you don’t have to copy and paste your content into another system.

Image analysis

Gemini incorporates Google Lens capabilities so you can upload images and text prompts. You can use the image to add context to your prompt or direct Gemini to do something with it.

You can use the image analysis functionality to perform a variety of tasks, such as:

• Get a description of what’s in an image.
• Write a caption for an image in a particular style or a particular length.
• Identify what’s pictured, like a specific flower or type of insect.
• Transcribe handwritten notes.
• Turn images of text, like your car’s vehicle identification number (VIN), into text.

One limitation of Gemini’s image features is that they don’t allow you to upload photos of people. This rule prevents people from using the platform to generate harmful images of others.

Image generation

Google Gemini can generate images based on your prompts. You can also ask Gemini to use a picture you upload as a reference or an inspiration. It’s capable of generating images in any style. For example, you can specify if you want your image to look photorealistic, abstract, hand-drawn, or like an oil painting.

Here are some ways you can use the image generation feature:

• Creating images for social media, presentations, and websites
• Drafting concept art for film, art, photography, or sculpture projects
• Adding illustrations to existing prose or poetry
• Creating your own library of stock images
• Re-creating an existing image in a different style
• Brainstorming ideas for decor

Code writing

Gemini can translate plain language instructions into code. It writes code in more than 20 programming languages.

Some of its coding capabilities include:

• Finding bugs, syntax errors, and logical errors in existing code
• Modernizing existing code
• Explaining the functionality of a snippet of code
• Creating documentation
• Translating code between different programming languages

Brainstorming

Gemini can assist you in generating ideas for creative projects, activities, and marketing campaigns.

You can ask Gemini to help you brainstorm for many activities:

• Ideas for fun games for a team-building, networking, or family event
• Features and functionalities for a product or service
• Layouts for visuals to accompany presentations, blog posts, or social media
• Prompts to use during brainstorming sessions
• Content for blogs, presentations, social media posts, and email campaigns
• New activities or hobbies to try based on your current interests and skills

Searching the internet

Gemini’s ability to leverage Google’s search capabilities is one thing that sets it apart. These capabilities can be used to search directly inside the application or to perform more complex tasks.

For searching the internet, it’s important to note that Gemini doesn’t produce results like what you would see on a Google search page. Instead, it summarizes them.

Sometimes, Gemini’s responses include images with links. So if you search for “major holidays in Kenya,” Gemini may respond with a list of holidays and images of people celebrating them.

You can add Gemini to Google search pages with a web browser extension. With the extension, you get a summary of the search page results. You can also prompt Gemini to do things with your search results. For example, if you’re trying to decide which television to buy, Gemini can create a comparison table so you don’t have to hop between tabs.

Interacting with Google apps and services

With Gemini Extensions, you can search Google’s many other apps and services: Gmail, Flights, YouTube, Docs, Drive, and Maps.

Here are some ways you can use this functionality:

• Find out when you last emailed a former colleague and get a summary of what you discussed.
• Find out the ingredients and measurements listed in a YouTube cooking video.
• Get a list of attractions in a city you plan to visit, with distance and average driving time from your hotel.
• Generate content ideas based on the topics discussed in a Google Doc.

You can also use Gemini inside Gmail, Docs, and Drive if you have the paid version of Gemini.

Summarize text

Gemini can scan texts and summarize them for you. You can paste any text or URL into the chatbot.

You can use this feature to do the following:

• Summarize an article with key points of interest for readers with a technical background.
• Pull out the most important topics from a transcription of an interview.
• Compare two articles with a high-level overview of them in an easy-to-read table.

Advantages of Gemini

Gemini offers several advantages that leverage Google’s extensive technology and information ecosystem, such as integrations with Google’s services, up-to-date information, and multimodal interaction.

Integration with Google products

Searching Google Flights, Maps, Hotels, Docs, and Drive within a single interface can have its advantages. For example, you can manage projects requiring multiple tabs, like planning an event, in a single view.

Here are more examples of how Gemini’s integration with Google can aid you in your workflow:

• Use the “Google it” feature to verify Gemini’s responses in real time.
• Dive deeper into your research by visiting links in the interface.
• Export Gemini’s responses directly to Gmail or Google Docs.

Real-time updates and recent information

Since Gemini pulls data directly from Google, it can incorporate timely information in its response.

Given these capabilities, you can ask Gemini about current events and topics:

• Create an image inspired by today’s weather in your city.
• Request a summary of the latest news in your country.
• Research current trends on topics that evolve quickly, like pop culture and technology.
• Find out which new laws were passed in the last year.
• Get updated guidelines from authorities like the Centers for Disease Control and the Federal Trade Commission.
• Find out who the current elected officials are in a municipality, state, or country.

Multimodality in a single platform

Google Gemini is multimodal, so it can read and generate code, text, images, and audio within a single application.

Multimodal capabilities offer many benefits:

• Greater context for prompts, which allows Gemini to understand nuances like humor or sarcasm that may be missed with text-only prompts
• More natural interactions with the platform, since you can tell it to look at an image or watch a video instead of trying to describe it yourself
• Multistep prompts, such as asking Gemini to write a social media post and create the accompanying image

Disadvantages of Gemini

Gemini, like all generative AI tools, has its disadvantages. These pitfalls can cause you to make errors, slow down your productivity, or use Gemini only for specific tasks.

Inaccuracies

Gemini may produce inaccurate responses. In the AI world, these are known as hallucinations. Since generative AI tools work by making predictions, it’s possible that sometimes these predictions will be incorrect. This means that a tool like Gemini can make errors even when summarizing information directly from the web. The sources it provides can be unreliable, so it’s a good idea to double-check them as well.

Gemini can even be inaccurate about its capabilities. For example, it may say it can’t create images or search the web. However, if you reword your prompt, it will then perform the task it said it couldn’t do.

Biases

Gemini can generate biased responses. In some cases, biases are caused by a lack of data, such as limitations around answers having to do with certain cultures or countries. Gemini is not alone in this problem—other generative AI tools show bias, too, because of gaps in their training data.

In other cases, biases are caused by negative stereotypes, discriminatory ideas, and political opinions from its training dataset. For instance, Gemini’s responses may include language implying favoritism for one side over another in an international conflict. Even though it’s not supposed to incorporate a point of view in its responses, these biases can still seep through.

Limited creativity

Though Gemini can generate creative content, it performs better for research tasks. Since Google is primarily known as an information provider, it makes sense that its chatbot favors more direct, informational responses.

For creative tasks, you may have to write highly prescriptive prompts and refine Gemini’s responses with follow-ups. You may even prefer other generative AI chatbots that were trained to generate more imaginative outputs.

Google Gemini and generative AI are constantly changing

Gemini is in a state of rapid change. Many experts say harnessing Google’s existing capabilities with sophisticated, conversational AI will change the face of search. Gemini can certainly change how you interact with Google apps and services today.

While Gemini unlocks new capabilities that help you be more informed and productive, it can also provide inaccurate, biased responses. Since generative AI is unfolding right before us, keeping up with the latest developments will help you maximize its benefits while minimizing its downsides.

The post Google Gemini 101: What It Is and How to Use It appeared first on Grammarly Blog.

Paraphrasing a paragraph is different than paraphrasing a sentence or phrase. In some ways, it’s more difficult, but in others, it’s easier. Knowing how to paraphrase a paragraph takes more than just changing a few words; you need to rewrite multiple sentences and understand the rules for citation, syntax, and avoiding plagiarism.

In this guide, we explain everything you need to know about paraphrasing a paragraph. We discuss steps, tips, and techniques, plus we share an example to illustrate exactly how to paraphrase a paragraph.

What does paraphrasing a paragraph entail?

Paraphrasing a paragraph is rewriting the ideas from a paragraph written or said by someone else in your own words and style. The points remain the same, but you restate the meaning with new, original writing.

When you want to use another person’s ideas in your own work, you typically have two choices: direct quotes or paraphrasing. Direct quotes work best if you like the original wording, but if you use too many quotations, your writing may become hard to read. Academic writing flows more naturally when you strike a balance between paraphrasing and direct quotes. Alternating paraphrases and quotations is an essential part of how to write a research paper. Keep in mind that both paraphrases and direct quotes require citations.

5 steps for how to paraphrase a paragraph

1 Read the source paragraph thoroughly

Don’t jump into rewriting a paragraph. First, take time to review the original carefully. Read and reread the paragraph until you’re sure you understand each part. Otherwise, you may misinterpret something in your paraphrase, or forget to include it completely.

2 Identify the key points, words, and ideas

Once you’re familiar with the source paragraph, consider which parts—whether it’s abstract ideas or specific terminology — you want to include in your rewrite. If your topic is slightly different from the source topic, only include the parts relevant to what you’re writing about.

3 Rewrite the paragraph from memory using different vocabulary and syntax

Replacing words with synonyms is a main strategy for paraphrasing, but it’s not sufficient on its own. You also need to change the syntax—the order that words come in. Because paraphrasing usually involves moving around a lot of words, be careful that you don’t inadvertently change the original meaning.

One advantage of paraphrasing a paragraph instead of a sentence is that you can sometimes change the order of the sentences. Paragraph structure typically uses a topic sentence (introduction) and a conclusion sentence, but the sentences in the middle can be rearranged except where the order matters, as with step-by-step instructions or logical progressions.

You can set apart your version even more by combining or splitting up some sentences. Look for clauses that can become standalone sentences or that you can attach to a different sentence.

Get rid of phrases or even entire sentences that aren’t related to what you’re writing about. Editing your paraphrases like this also makes your writing stronger by focusing only on your topic. If you plan on removing parts, paraphrasing a paragraph makes more sense than quoting it, because this lets you avoid using ellipsis breaks, which makes some quotes hard to read.

4 Review to ensure you restate the meaning correctly

When you move and change words, you may inadvertently introduce technical missteps, including grammar and spelling mistakes. Use Grammarly to check your paraphrase for clarity, conciseness, and correctness.

5 Use a citation to avoid plagiarism

Even though you’re using original words and your own writing style, you still need to include a citation. That’s because the phrasing may be your own, but the ideas are not.

Citing paraphrased text usually involves parenthetical citations, a type of in-text citation that puts the author’s last name in parentheses, along with the page number, year of publication, or both. Different formatting styles, such as APA, MLA, and Chicago, all have different rules for citations, so check our guides for the one you’re using.

In addition to the in-text citation, you also need to include a full citation at the end of your work, in the bibliography section. The formatting for your full citation also depends on the formatting style. Use Grammarly’s citations generator to help you instantly add citations to your papers.

Paraphrasing a paragraph example

Original paragraph

A human being is a part of the whole, called by us “Universe”, a part limited in time and space. He experiences himself, his thoughts and feelings as something separated from the rest — a kind of optical delusion of his consciousness. This delusion is a kind of prison for us, restricting us to our personal desires and to affection for a few persons nearest to us. Our task must be to free ourselves from this prison by widening our circle of compassion to embrace all living creatures and the whole of nature in its beauty. Nobody is able to achieve this completely, but the striving for such achievement is in itself a part of the liberation and a foundation for inner security.

—Albert Einstein,Condolence letter to Rabbi Robert Marcus over the loss of his daughter

Paraphrased version

What we call the “universe” connects everyone, but our individual experiences are still separated from each other. However, this sense of isolation is just an illusion that limits our worldview to only what is familiar to us. We can dispel this illusion by showing empathy to all humans, animals, and nature as a whole, even what is new and unfamiliar. This may be a lofty or even impossible goal, but even just attempting it is a good start to embracing the unity of the universe and, consequently, finding inner peace. (Einstein, 1950).

Tips and techniques for paraphrasing a paragraph

We’ve covered the most effective techniques for paraphrasing a paragraph. As a reminder, they’re summarized below.

• Add or remove sections—You can make an idea unique by adding new parts that weren’t in the original or removing parts that aren’t related to your topic.
• Replace words with synonyms—When people think of paraphrasing, they usually think about using synonyms, different words with the same meaning.
• Switch the part of speech—A part of speech (or “word class“) refers to the function of a word, like a noun, verb, adjective, etc. Changing the part of speech, such as turning a verb into a noun, is a great way to rewrite an idea without losing its meaning.
• Change the sentence structure—In addition to moving parts of the paragraph, you can also move parts of individual sentences, like swapping the position of phrases or using a different subject.

While these techniques have been proven effective, they still take time to learn. Keep in mind that Grammarly’s free paraphrasing tool can give you a few different recommended paraphrases. And once you’ve got a draft, you can rewrite with AI to polish it.

Common Paraphrasing Mistakes

• Not altering the words sufficiently
• Changing the meaning
• Forgetting to cite the source

Paraphrasing a Paragraph FAQs

How much should I change when paraphrasing a paragraph?

Ideally, you want to change as much as you can when paraphrasing a paragraph. More than just individual words, try to rethink the entire paragraph as a whole and create a whole new structure. Aim to restate the meaning in your own style to stay consistent with the rest of your writing.

What are some techniques to help rewrite paragraphs?

The most common paraphrasing technique is to replace some words with synonyms, but this is not enough on its own. It’s best to rearrange the order of sentences, as well as combining and separating some of the originals. You can also add new ideas, or get rid of some that aren’t related to your topic.

Do you need citations when paraphrasing a paragraph?

Even though you’re changing the words, you also need a citation when paraphrasing a paragraph. Paraphrases usually use a parenthetical citation within the text and a full citation in the bibliography at the end.

The post Paraphrasing a Paragraph: An Easy Step-by-Step Guide appeared first on Grammarly Blog.

Does the term “public speaking” make your palms sweat or heart race? While it remains a top source of anxiety for many, public speaking is an inescapable part of thriving in the workplace. After all, effectively talking to others is essential to communicating your ideas, vouch for your initiatives, or leading a team. Speaking publicly at work doesn’t necessarily mean getting up on a stage in front of thousands of people. It could mean speaking to your teammates about a project or briefing leadership on an initiative.

Whatever the forum, public speaking is always easier when you prepare a presentation and talking points ahead of time. Knowing how to write for public speaking can make the task feel less daunting and increase the chances you’ll deliver a knockout speech. This guide will cover how to write presentations for various audiences you work with every day.

What is writing for public speaking?

Writing for public speaking is an important part of most people’s careers, whether they give presentations to their team for internal communications or deliver speeches at major industry events. It involves writing your key talking points, drafting a script for a speech, or creating visuals to accompany your spoken presentation.

Writing for public speaking is different from other types of business writing because your audience will listen to instead of read your content, which requires special attention to delivery, timing, clarity, personal engagement, and visual aids.

Great speeches take an enraptured audience with them on their journeys.

Foundations of effective writing for public speaking

Engaging presentation content starts with effective writing for public speaking. Make sure your script or talking points have the following qualities:

Clarity and conciseness

When language is clear and concise, your speech will be easier to follow and, thus, more likely to communicate your intended message. To achieve clarity, use shorter and simpler words and sentences to convey your point. Be direct. For example, rather than saying, “Our recent external audit of public perception of our most recent product demonstrated a generally positive sentiment,” try: “People generally liked our new product.”

Human interest and storytelling

Storytelling always makes speeches more engaging. Engaging speeches take the audience on a journey. Instead of focusing solely on facts and figures, include elements that tug on people’s emotions or needs in order to engage your audience the entire time. Pitching a product? Tell a story about how a customer benefitted. Recapping the results of a campaign? Highlight the teamwork that made it possible and take care to include those who contributed. Anecdotes, personality, and audience interaction can help your content become more memorable and will make the presentation more delightful and impactful.

Focused points

When you ramble or go off on tangents, you’re more likely to lose your audience’s attention. Focus your speech on a few relevant points or characters. Don’t overload your audience with too much information. If your goal is to fill your team in on the progress of an initiative, don’t talk about the five other things the company is doing. If you’re speaking about the benefits of your company, don’t spend too much time on what your competitors are doing. Make sure you stay focused on your main points. You may find it helpful to reiterate those points at the end of your presentation for extra impact.

Writing for Different Public Speaking Scenarios

In a business setting, you may find yourself in many situations in which you’ll need to speak to an audience, requiring you to tailor the tone and style of your presentation.

Presentations to a large group

When presenting to a large group, make sure that the content of your presentation is accessible to all members of the group, especially the least expert ones. Don’t use too many acronyms and avoid jargon they may not know. Leave time for questions at the end or even during your presentation. Pausing your monologue to engage the audience or check for comprehension makes it more likely that you’ll capture and retain their attention.

Team meetings

When speaking to a team, your goal may be to foster collaboration, generate ideas, promote morale—or all three. As such, these kinds of speeches tend to be more conversational and collaborative. Come prepared with an agenda or talking points, and leave a polished script for another occasion. Incorporate pauses in your presentation to take questions. If you’re asking your team to do something, such as brainstorm, make sure you are clear about that direction by displaying a guiding question as a visual aid. Lastly, balance informational content with motivational messages since a large part of team meetings is to generate excitement within your team.

Leadership briefings

Leadership briefings should be direct and focused on impactful information. Leaders have limited amounts of time and many prefer to have information presented to them in bite-sized chunks. Focus on key results and impact on the business. Executives also tend to like visuals that are easy to understand and scannable. Instead of loading lots of text on a slide, include charts and graphs that illustrate results, and present information in bullet points and tables.

Company announcements

Company announcements tend to be more formal than team meetings, so you might come with more thorough notes or scripting that’s even been vetted by human resources, legal departments, and other stakeholders. When delivering a speech on a company announcement, write down a few key messages to deliver. If it’s a quarterly review, for example, highlight the story of the quarter: Where the company started out, where it progressed, and any notable hiccups or success stories along the way. If you’re delivering sensitive news, do so directly, with empathy, and use pre-approved language.

Tips for more impactful public speaking delivery

Write for your audience

The key to public speaking success is to use your audience’s needs as the guiding force for your writing. Putting yourself in the audience’s shoes builds empathy and connection with them. Center their needs, interests, and expectations and write accordingly to help ensure your message lands as you intend. When communicating with large or varied audiences, keep your words simple, short, and inclusive. While you may be speaking to a large crowd, it helps to write as if you’re speaking to just one person so it comes across as personal, conversational, and engaging.

Come with notes—not a script

When just starting out writing for public speaking, it may be tempting to write down what you’ll say word-for-word. That may be fine if you’re giving a formal speech and need to have a very specific script. However, for most presentations, such as team meetings, it’s better to write an outline or a list of points you’d like to address. That way, you can look down at your notes to make sure you’re on track in the presentation, but won’t sound as stiff or robotic as you would if reading off a sheet of paper.

Revise and refine your written content

As with all content, editing your writing helps ensure it’s clear and impactful. Read it out loud to make sure it sounds natural and engaging. Instead of focusing on spelling or grammar, use Grammarly’s writing suggestions to revise your content so it’s clear, impactful, and speaks to your objectives.

Incorporate visuals and multimedia

Visuals in a PowerPoint or Google Slide deck can enhance your presentation. They give the audience something to look at while you talk so that their attention doesn’t drift away. Visual aids such as videos, images, photos, charts, and graphs should be simple and not include large blocks of text.

Practice

Rehearsing always makes showtime easier and more successful. Try to reproduce the conditions you’ll face on the day of the speech. Stand up if you’ll be on a stage, and sit down at a table if you’re presenting in a meeting room. If you’re presenting virtually, test your lighting, background, mic, video, and wifi. Speak slowly and clearly and make sure you know how to pronounce all the words you’ve written. Practicing out loud gives you a chance to make sure your points are coherent and that you’ll make sense to a live audience.

Ask for feedback

It’s worth it for important presentations to ask for feedback from a trusted friend as you practice. If that’s not available, record yourself giving a presentation and watch it back. Ask yourself: Did the points you wanted to say come off well? Did you speak loud enough? Did you use your hands enough? Too much? Grading yourself while you practice allows you to make any tweaks necessary before the big day.

Key points

• Writing for effective public speaking must be concise, clear, and tell an engaging story so that the presentation holds the audience’s attention.
• Effective public speaking can have concrete business benefits. It can amplify your team’s morale and productivity, convince investors to invest in your company, or show your superiors that an initiative you championed was successful.
• Use presentations and visual aids to your advantage by incorporating charts, graphs, tables, and bulleted messages.
• Come with notes on the main points you’d like to address in a presentation, rather than a script to read word-for-word. Doing so will make your speaking seem more relaxed and natural.
• In order to improve your writing for public speaking, practice your presentation and ask for feedback to ensure your speech is the most impactful it can be.

Writing for public speaking FAQs

How can I adapt my writing for diverse audiences?

Consider who is in your audience, their level of understanding of the topic you’re discussing, and how they prefer to receive information. When speaking to another team about your team’s project, you may need to give them more context about the project before delving into the details. When speaking to executives, focus your content on digestible, easy-to-understand results.

What are the key differences between writing for in-person versus virtual presentations?

When writing for in-person presentations, you can rely more on your body language and audience energy, like whether they are laughing or not, to tailor your speech. When writing for virtual presentations, it’s harder to communicate with your entire body and to gain immediate audience feedback. In a virtual setting, you might use tech features, such as the chat section, to engage the audience or field questions.

How do I balance technical information with engaging content?

Try to use technical information only when necessary. Ideally, your full presentation will consist of engaging content, even if it’s technical. Focus on the key takeaways and simplify technical information to explain it to your audience concisely. Use visual aids or videos to help with understanding. Starting with a human interest anecdote and summarizing the key takeaways will make your presentation more engaging.

What techniques can help ensure my main points are memorable?

You can ensure your main points are memorable by pulling out the most impressive data points about the project you’re presenting on, and reiterating them at the end so the audience remembers them.

The post The Art of Writing for Public Speaking: Crafting Presentations for Impactful Delivery appeared first on Grammarly Blog.

You’re writing a note to a colleague asking to have a “quick sync” to make a “game-time decision” on a “rock star” candidate you’re sure will “give 110 percent” to the job.

Few things are as potentially confusing (or grating) as jargon in business writing. It’s true that jargon can sometimes be a convenient shorthand when you’re communicating quickly to someone you’re pretty sure will understand you. But even if they do, it runs the risk of making your writing seem like in-group technobabble. For this and other reasons, you should avoid using jargon phrases in business writing.

What is jargon?

Jargon is a term or phrase used by a specific group or discipline. The words in a jargon phrase often convey an idea different from the words’ literal meaning. For example, “giving 110 percent” has nothing to do with math and is figurative language for “working very hard.”

Jargon is the opposite of plain language, which is direct, accessible language that is designed to communicate an idea simply. Jargon differs from industry-specific terminology because it is a way of saying something that can be said in a simpler, more direct way.

Jargon in business communication

Jargon can appear anywhere in business communication: a formal email to a client, an internal project proposal, the executive summary of a white paper, or elsewhere. Using jargon can be a colorful way to communicate a concept. But more often, jargon can make your writing appear vague and pretentious.

Jargon can make you appear like you’re trying to impress others, which has the counterintuitive effect of making you appear less skilled. In fact, a study by Harvard Business Review found that academics who were less experienced used more jargon than those who were more experienced. It’s often those who lack real expertise who use jargon to give the impression that they know what they’re talking about.

Regardless of your own knowledge, jargon can alienate or confuse others who do not understand the figurative phrase, especially those who aren’t native speakers of English and—given the plethora of sports phrases in business jargon—those who aren’t sports fans. If someone doesn’t understand your writing, you’ve failed to communicate with them. Jargon sometimes conceals an insensitive attitude or comes off as thoughtless, trendy, or clichéd. Lastly, jargon can be used to obscure or sugarcoat a less-than-pleasant idea, which can make it look like you’re avoiding a difficult subject. Think of the phrase “reduction in workforce” when a company needs to lay off a bunch of people.

For all of these negative reasons, jargon is unnecessary. If your idea can be conveyed more simply and accessibly without using jargon, it should be. You’ll seldom go wrong speaking clearly and directly.

Five common jargon phrases

Here are common jargon phrases frequently used in business writing:

Think outside the box

This phrase is vague and unclear. It suggests you’re asking someone to come up with new ideas but doesn’t specify the parameters of that request or what kind of ideas you’d like. It doesn’t tell them what, exactly, the box is. What’s more, it’s been overused and has lost whatever impact it once had. So it’s hard for people to understand or respond to this call to action. A clearer directive would be “come up with three new taglines” or “brainstorm four new ways our team can work with this other one on a weekly basis.”

Low-hanging fruit

This expression refers to something that is easy to achieve, as the orange that hangs on a low branch is easy to pluck. It’s relatively clear what it means, but it has a problematic association, and like the other expressions discussed here, it now lacks impact. If what you mean is “easy to achieve” or “a quick win,” why not say so?

Up the ante

This means to make something better or to put more effort into something, but it is unclear how. Do we put in more hours? Rethink a portion of the proposal? Or collaborate with another team we haven’t worked with yet?

Monday-morning quarterback

This refers to someone who watches a game on Sunday and decides the next day how it should’ve been played. In business use, it refers to someone who second-guesses a decision that they or someone else made in the office. Like other sports jargon phrases, this can be unclear, especially to the many colleagues you have who are not fans of American football.

Out of pocket

This means you’ll be unavailable. But it’s not clear how it came to mean this, or what pockets have to do with your schedule in the workplace. A better way to say this would be to more directly state that you will not be reachable by email or Slack—or just that you’ll be out of the office, if that’s the case.

Technical terms vs. jargon: Understanding the difference

Technical terms are specialized words or phrases that refer to something specific in a discipline. They are different from jargon because they are necessary phrases to communicate a very specific idea, which can’t readily be paraphrased into simpler words. One example of a technical term in marketing is “sales-qualified lead” or “SQL.” This is a term that refers to someone who has expressed interest in your company (a “lead”) that the sales team has agreed is a good prospect (“qualified”). There’s not really an easier, industry-approved way to say this.

Jargon is different from technical terms because it often is a more conversational phrase or expression that can be expressed in another, easier way. Saying someone is a “rock star at generating SQLs” is jargon. You could instead say that that individual is excellent at identifying people who are interested in your product.

Is jargon ever appropriate?

Jargon is sometimes appropriate in certain contexts. One example is when you’re talking with people who have used those phrases in the past, since you can be pretty sure they understand the phrases. (Of course, there’s a chance that they did not understand the jargon themselves when they used it!) Another way jargon can be appropriate is to serve as convenient shorthand when you and your audience already know what’s meant. For instance, in an ongoing discussion, once it’s clear what “disintermediation” involves, you may save words by relying on that term.

Role of audience in business writing

Your audience impacts all aspects of business writing, from the tone you take to the words you choose to use. Jargon can be appropriate to use when you’re communicating with peers whom you can be reasonably sure will understand the phrases. But if you’re talking to a wider audience that includes many people of different levels, it’s best to err on the safe side and avoid jargon in business writing.

Improve clarity and readability in business documents

It’s essential for business writing—regardless of whether you’re writing an email or a formal proposal—to be clear and direct so that the reader will understand your intended message. The best way to ensure this is to use plain language as opposed to jargon. Here are some tips for doing that:

Avoid abbreviations

One of the simplest rules is to write out the abbreviation rather than using the letters. For example, write out “marketing-qualified lead” rather than “MQL.”

Paraphrase and simplify

Challenge yourself to say more simply, by paraphrasing, the jargon you’re tempted to use. Pretend you’re speaking to a child. So instead of saying, “The juice is not worth the squeeze,” say, “It’s not worth it.”

Look it up, then replace it

If you’re not sure of a phrase’s exact meaning, don’t use it. Rather, look up the phrase you’re tempted to use to see what it means. Then, use an alternate phrase that more clearly conveys what you really want to say.

Assume the reader has no knowledge of your area

This assumption encourages you to use clear and simple words rather than words or phrases that you may hear tossed around.

Key takeaways

• Jargon is a term or phrase that means something other than the literal meaning of the word or words.
• Using jargon in business writing can make you come off as pretentious, exclusionary, and insecure.
• Using jargon can also confuse the person you’re giving guidance to. They may not understand what you’d like them to do.
• Writers should always prioritize clarity and simplicity for effective communication and should try to avoid jargon in most instances when they are writing in the workplace.

Jargon FAQs

Can jargon ever be appropriate in business writing?

Yes, jargon can sometimes be appropriate in business writing if you’re pretty sure the reader understands the expression you’re using or if it can serve as convenient shorthand for something you’ve already clarified. However, it should still be limited since too much jargon in one written piece can irritate the reader.

How do I know if a term is jargon or necessary technical language?

One way to tell if a term is jargon or necessary technical language is to try to paraphrase it in simpler terms. If it’s easy to do so, then it’s probably jargon, and you should go with the simpler phrase.

What are the best strategies for simplifying jargon-heavy writing?

The best strategies for simplifying jargon-heavy writing are to rephrase simply what you’re trying to say, assume the audience has no experience in your discipline, and avoid using abbreviations.

How can I make my business writing more accessible to a general audience?

Business writing should be clear and concise, and it should use simple words whenever possible. You can also assume that the audience who will read your writing has no expertise in your industry or area and that your job is to explain it to them as clearly as possible.

The post Jargon Phrases to Avoid in Business Writing appeared first on Grammarly Blog.

When you think about generative AI models, you probably think about the large language models (LLMs) that have made such a splash in recent years. However, generative AI itself dates back many decades, and LLMs are just the latest evolution. And alongside LLMs, many different kinds of generative AI models are used for different generative AI tools and use cases, such as diffusion models that are used for image generation.

In this article, we’ll explain what generative AI models are, how they’re developed, and provide a deeper dive into some of the most common generative AI models today—enough to give you a conceptual understanding of these models that will impress your friends and colleagues, without you needing to take a college course in machine learning.

Table of contents

• What is a generative AI model?
• How generative AI models work
• How are generative AI models developed?
• Types of generative AI models
• Conclusion

What is a generative AI model?

Generative AI models are a subset of artificial intelligence systems that specialize in creating new, original content that mirrors the characteristics of their training data. Through learning from patterns and relationships in data, these models can generate outputs like text, images, sounds, or videos that resemble the style, tone, and nuances of their source material. This capability positions generative AI at the heart of innovation, allowing for creative and dynamic applications across diverse fields by interpreting and transforming input data into novel creations.

How generative AI models work

Generative AI models function by leveraging a sophisticated form of machine learning algorithm known as a neural network. A neural network comprises multiple layers of interconnected nodes, each represented by a snippet of computer code. These nodes perform minor, individual tasks but collectively contribute to making complex decisions, mirroring the neuron functionality in the human brain.

To illustrate, consider a neural network tasked with distinguishing between images of pies and cakes. The network analyzes the image at a granular level, breaking it into pixels. At a very basic level, there will be different nodes in the network dedicated to understanding different pixels and groups of pixels. Maybe some will look at whether there are layers in the dessert, while others will determine if there’s frosting or a crust. The nodes each store information about the features of what pie vs. cake looks like, and whenever a new image comes into play, it’s processed through each and every node to output a final prediction.

In the context of generative AI, this principle extends beyond simply recognition to the creation of new, original content. Instead of merely identifying features, generative models use neural networks to understand the underlying patterns and structures of the data they’re trained on. This process involves complex interactions and adjustments within the neural network, guided by algorithms designed to optimize the creativity and accuracy of the generated output.

How are generative AI models developed?

The development of generative AI models involves a series of complex and interrelated steps, typically carried out by teams of researchers and engineers. These models, such as GPT (generative pre-trained transformer), from OpenAI, and other similar architectures, are designed to generate new content that mimics the distribution of the data they were trained on.

Here’s a step-by-step breakdown of that process:

1 Data collection

Data scientists and engineers first determine the goals and requirements of their project, which guides them to collect a wide and appropriate dataset. They often use public datasets, which offer vast quantities of text or images for their needs. For instance, the training of ChatGPT (GPT-3.5) involved processing 570GB of data, equivalent to 300 billion words from public internet sources, including nearly all of Wikipedia’s content.

2 Model selection

Choosing the right model architecture is a critical step in developing generative AI systems. The decision is guided by the nature of the task at hand, the type of data available, the desired quality of the output, and computational constraints. Specific architectures, including VAEs, GANs, and transformer-based and diffusion models, will be discussed in more detail later in this article. At this stage, it’s important to understand that new models often start from a preexisting architecture framework. This approach leverages proven structures as a foundation, allowing for refinements and innovations tailored to the unique requirements of the project at hand.

3 Model training

The chosen model is trained using the collected dataset from the first step. Training generative AI models often requires a large amount of computing power, using special hardware like GPUs (graphics processing units) and TPUs (tensor processing units). While the training approach varies based on the model architecture, all models go through a process called hyperparameter tuning. This is where data scientists adjust certain performance settings to achieve the best results.

4 Evaluation and fine-tuning

Finally, model performance is evaluated or tested in the real world. Evaluating generative AI models is a bit different from evaluating traditional machine learning models because generative AI creates an entirely new output, and the quality of this output tends to be subjective. Metrics differ based on what the model is creating, and evaluation techniques for generative AI typically include using human raters—and may employ the strategy of having generative AI models evaluate one another. Learnings from the evaluation stage are typically applied back into fine-tuning the model or even retraining it. After the model’s performance is validated, it’s ready for production.

Types of generative AI models

Building on our foundational knowledge of generative AI models and the neural networks that power them, we’re now set to dive into specific types of model architectures that have emerged since the early 2010s. We’ll explore each model’s unique strengths and weaknesses, as well as their practical applications.

Here’s a brief overview of the models we’ll be discussing:

• Variational autoencoders (VAEs) are adept at learning complex data distributions and are often used for tasks like image generating and editing.
• Generative adversarial networks (GANs) are known for their ability to create highly realistic images and have become popular in a variety of creative applications.
• Diffusion models are a newer class of models that generate high-quality samples through a process of gradually adding and then removing noise.
• Language models excel at understanding and generating human language, making them useful for applications like chatbots and text completion.
• Transformer-based models were initially designed for natural language processing (NLP) tasks but have been adapted for use in generative models due to their powerful ability to handle sequential data.

Let’s delve deeper into each of these architectures to understand how they work and where they can be best applied.

Variational autoencoders (VAEs)

Variational autoencoders were invented by Max Welling and Diederik P. Kingma in 2013. They rely on the fact that a neural network can encode the high-level concepts the model learns during the training step. This is sometimes referred to as a “compression” or “projection” of the raw data.

If a model looks at an image of a cake, for example, it might turn that into an encoding containing all of the image’s features—sprinkles, frosting color, spongy layers, etc. This encoding looks like a set of numbers that makes sense to the model but not to humans. It can be decoded by yet another neural network to try to re-create the original image—though it will have some gaps because the encoding is a compression. This type of model, with the encoder and decoder pieces working together, is called an autoencoder.

Variational autoencoders put a spin on the autoencoder idea to generate new outputs. When generating its encodings, a VAE uses probabilities instead of discrete numbers. After all, does whipped cream count as frosting? Sometimes yes; sometimes no.

It turns out that if you train a neural network to create these probabilistic encodings and train another neural network to decode them, you can get some pretty interesting results. The decoder can sample points in the variational encoding “space” and create entirely new outputs that will still appear realistic because they have preserved the probabilistic relationships of the training data.

Advantages and disadvantages

Variational autoencoders use unsupervised learning, which means that the model learns on its own from raw data without requiring humans to label different features or outcomes. Such models are especially successful at creating content that deviates slightly from the original. Because of how they work with encodings, they can also be given specific instructions based on features of the training data: “Show me a dessert that represents the perfect midpoint between cake and pie.” That said, VAEs optimize for likely outcomes, so they are unlikely to excel at creating very original or groundbreaking content.

One common complaint about VAEs is that they can produce noisy (i.e., blurry) images due to the fact that encoding and decoding involves compression, which introduces loss of information.

Use cases

Variational autoencoders work with all kinds of data, though they are primarily used to generate images, audio, and text. One interesting application is anomaly detection: In a dataset, VAEs can find the data points that deviate the most from the norm, because those points will have the highest reconstruction error—meaning they will be the furthest from the probabilities that the VAE has encoded.

Generative adversarial networks (GANs)

Generative adversarial networks were developed by Ian Goodfellow in 2014. While neural networks had been able to generate images before that, the results tended to be blurry and unconvincing. The core question (and insight) behind GANs is this: What happens if you pit two neural networks against each other? One, called the generator, is taught to generate new content, while another, called the discriminator, is trained to know the difference between real and fake content.

The generator creates candidate images and shows them to the discriminator. Based on the feedback, the generator updates its predictions accordingly, getting better and better at “fooling” the discriminator. Once it can fool the discriminator 50% of the time (as good as a coin toss between real and fake), the feedback training loop stops. The generator part of the GAN is then ready for evaluation and production.

Since 2014, hundreds of variations of GANs have been developed for different use cases and to balance the inherent advantages and disadvantages of GANs.

Advantages and disadvantages

Generative adversarial networks, along with VAEs, initially sparked a lot of buzz around the potential of generative AI. They use unsupervised learning, so the model gets better on its own without researchers needing to tell it whether its outputs are good or bad. Generative adversarial networks also manage to learn very quickly; compared to other existing solutions when they were first released, they could get good results with much less training data—hundreds of images compared to thousands.

However, GANs generally struggle to create content that doesn’t resemble their training data—they are impersonators, not creators. And sometimes they can “overfit” their training data, such as when GANs created images of cat photos containing letters because they were shown a lot of cat memes.

Training a GAN is a challenge. Two networks must be juggled during training. Issues can also arise when the discriminator is too good, leading to training cycles that never end—or if the discriminator is not good enough, which leads to poor outcomes. They can also suffer from what’s called mode collapse, where they fail to produce diverse outputs because the generator learns a few ways to trick the discriminator and focuses on those strategies to the exclusion of others.

Use cases

Generative adversarial networks are used primarily to generate content that is very similar to the original. For example, they can produce convincing human faces or realistic photos of interiors or landscapes for use in stock photography or video games. They can also create images that have been altered in some way, such as changing an image from color to black and white or aging a face in an image. That said, not all GANs produce images. For example, some GANs have been used to produce text-to-speech output.

Diffusion models

Diffusion models also came about in the mid-2010s, offering some breakthroughs that delivered better performance by the early 2020s. They power image-generation tools like DALL-E, Stable Diffusion, and Midjourney.

Diffusion models work by introducing Gaussian noise to an image, distorting it in a series of steps, and then training a model to reverse these steps and transform the “noisy” image into a clear one. (“Gaussian noise” just means the noise is randomly added using a bell curve of probabilities.)

You can think of the noisy image as being kind of like the VAE encoding, and indeed VAEs and diffusion models are related. Training-data images of, say, key lime pie, will end up with pretty similar noisy versions. But even the same noisy image won’t be “denoised” to the same thing every time, because the model is making educated guesses along the way.

You might have already figured out where the generative part comes in. If you give the model a representation of the image in the noisy space, it will be able to denoise the image and come up with an entirely new, clear picture. It’s sort of like how the decoder samples from the encoding. But there’s one important difference: There hasn’t been any compression along the way. So there’s been no real loss of data, and the resulting image will be higher-quality.

Generative AI tools that go from a text prompt to an image do that with the help of a separate model that understands how something like a “unicorn-themed birthday cake” might map to different image features. The noisy version of those features is then reversed to reveal a clear picture.

Advantages and disadvantages

Diffusion models don’t compress the training data, so they manage to create very realistic, high-quality images. However, they take significantly more resources and time to train than other models. That said, the training itself is more straightforward because they don’t run into the mode collapse of GANs and other drawbacks of the adversarial network. They also don’t suffer from the loss of data (and resulting lower-quality outputs) that VAEs have.

Use cases

Diffusion models are primarily used for image, sound, and video generation. There’s no inherent reason that they couldn’t be used to generate text as well, but so far, transformer-based models have been more effective for natural language.

Language models

Language models refer to any machine learning technique that generates a probabilistic model of natural language. The most well-known type of language model today is large language models (LLMs), which are trained on massive amounts of raw data and use a transformer-based architecture to generate text. (More on transformers in the next section.)

Before transformer-based models, most state-of-the-art language models used recurrent neural networks (RNNs). Recurrent neural networks introduce small loops in the interconnections between the nodes so that in addition to learning from the present signals, as in a traditional feedforward neural network, nodes can also learn from the recent past. This is important for processing or generating natural language, like a stream of text or a voice input. Unlike images, language is highly contextual—how we interpret it depends on what has come before.

Advantages and disadvantages

Because “language models” refers to such a large group of models, it’s difficult to generalize about their advantages and disadvantages. The challenges of language modeling include the fact that language is so high-dimensional—there are a vast number of different words in any given language, and some combinations might never appear in the training data.

Furthermore, language depends greatly on the context of what has come before in the sequence, requiring the network to handle or represent that context in some way. The capacity to address this need has led RNNs with long- and short-term memories and subsequently transformers, which can process an entire sentence as a whole, to emerge as the state-of-the-art architecture for language models.

Use cases

Language models can be used for translation, summarization, grammatical error correction, speech recognition, and many more tasks. They are used to generate new creative text content with many applications and are proving to be capable of advanced reasoning, such as analyzing data and solving logic puzzles. Interestingly, research has found that an emergent capability of LLMs is spatial awareness and the ability to create basic drawings, even though they are trained entirely on text.

Transformer-based models

Transformers, invented by researchers at Google and the University of Toronto in 2017, revolutionized the field of deep learning. Large language models like ChatGPT are transformer-based models, and Google search results are also powered by transformers.

A transformer-based model uses its training data to learn how different words are related. For example, it might learn that cake and pie are conceptually similar, whereas cake and cape are not directly related. It might also learn that slice can be linked to cake and pie, especially if those words occur in close proximity.

When analyzing text, the model uses this baseline understanding to construct what resembles a massive spreadsheet. It can look up any two words in the text and get an answer about how related they probably are.

By leveraging these contextual cues, a transformer model adeptly interprets language and forecasts potential continuities in a conversation. For instance, if someone mentions a cake in one segment and then shifts to discussing their birthday in the next, the model anticipates the eventual mention of candles or a party, based on the established linguistic connections.

Advantages and disadvantages

When it comes to analyzing and generating language, transformers have a few advantages over RNNS, their predecessors. They can process text in parallel across the network rather than processing each word sequentially. This makes them faster and more efficient to train on very large datasets. They can also make connections between words regardless of how far apart they are, allowing them to leverage more context from the text.

However, transformers need a lot of data to perform well, and with smaller datasets, more traditional neural network architectures may work better.

Use cases

Transformers have many generative AI applications. While transformer-based models are typically used to generate text or speech, researchers are exploring their use for image generation, as they are less computationally intensive than diffusion models.

Most famously, LLMs are transformer-based models. Language models use only the decoder portion of the architecture. The prompt is fed into the model as an encoding—that set of numerical values, probabilities, and attention data we mentioned earlier. The model decodes the input using the self-attention mechanism and by looking at all the words in the prompt in parallel. The model’s goal is to output a prediction for the next word in the sentence.

Transformers have many applications outside of generating text in natural language processing. In fact, they were originally conceived to translate, or transform, text from one language to another. Grammarly has contributed research toward using transformers to correct grammar mistakes.

Conclusion

Generative AI models have come a long way in the past decade. We hope that now you understand a little bit more about the evolution of these models, how they work, and how they might be applied to different use cases. This article has just scratched the surface, however, and left out many important details with the aim of providing an overview for the average reader. We encourage you to continue learning about the math and science behind these models by studying the research papers that they are based on and learning more about how they work from a probabilistic and statistical perspective.

The post Generative AI Models, Explained appeared first on Grammarly Blog.

Chatbots have emerged as one of the most common forms of generative AI. Claude AI is a chatbot built with ethics and safety in mind. Here’s an overview of what Claude AI is, how to use it, and which limitations you should be mindful of.

Table of contents

• What is Claude AI?
• How Claude AI works
• Who created Claude AI?
• Claude AI vs. ChatGPT
• Is Claude AI free to use?
• How to use Claude AI
• Advantages of Claude AI
• Disadvantages of Claude AI
• Conclusion

What is Claude AI?

Claude AI is an artificial intelligence chatbot. You can converse with Claude using natural language, just as you would with another person. Claude can generate various forms of text content such as summaries, creative works, and code. You can also upload images and text-based files to add context to your prompts.

Understanding the inner workings of Claude AI

Underlying the Claude AI chatbot is a large language model (LLM) that is also named Claude. An LLM is an AI model trained to recognize patterns and associations in large volumes of text. The model can then generate convincingly humanlike text responses.

The Claude LLM is based on the transformer architecture. Essentially, the transformer enables the model to make associations between words to understand context, meaning, and language patterns. The transformer architecture is also used in other popular generative AI tools such as OpenAI’s ChatGPT and Google Gemini.

Claude was trained with publicly available data from the internet, content licensed from third parties, and data provided by users and crowd workers for training.

Once the model processes the data, it uses a highly sophisticated set of probabilities to generate responses. Every response is a prediction of what the next word should be, like a souped-up form of autocomplete. Each word is predicted one at a time. It’s important to note that generative AI models by themselves don’t have knowledge in the same way we do. They have highly advanced algorithms that enable them to make predictions about what the right response should be.

Constitutional AI: Making Claude helpful, honest, and harmless

Harm reduction is key to Anthropic’s mission, which includes making models helpful, honest, and harmless. Anthropic’s approach to implementing guardrails differs from other AI tools. All generative AI tools must be fine-tuned to minimize harm with their responses and not engage with people who want to use them for malicious purposes. AI researchers typically address this problem by using humans to review multiple responses for a prompt and weed out the ones that are biased or profane, or spread misinformation or are toxic in other ways. Over time, the model learns which prompts to avoid and which responses are harmful.

Claude was fine-tuned not only with human feedback but also with a second model in a process called Constitutional AI. The logic behind Constitutional AI is that AI can be trained to moderate itself using a core set of principles. This is called reinforcement learning from AI feedback (RLAIF). One benefit of RLAIF is that it’s easier to define and adjust the guardrails. Also, as AI generates longer, more complex responses, it will be harder for human reviewers to keep up with the volume of information to assess. RLAIF, on the other hand, can scale easily.

Constitutional AI uses a set of principles derived from several sources, including the United Nations Universal Declaration of Human Rights. The principles are geared toward helping AI recognize toxic prompts, reduce biased responses, make clear distinctions between AI and humans, and reflect values that benefit humanity.

Some of the principles of Constitutional AI include:

• Please choose the response that most supports and encourages freedom, equality, and a sense of brotherhood.
• Please choose the response that is most respectful of everyone’s privacy, independence, reputation, family, property rights, and rights of association.
• Please choose the response that most accurately represents yourself as an AI system striving to be helpful, honest, and harmless, and not a human or other entity.
• Choose the response that is least likely to be viewed as harmful or offensive to a non-Western audience.
• Choose the response that is least likely to imply that you have preferences, feelings, opinions, or religious beliefs, or a human identity or life history, such as having a place of birth, relationships, family, memories, gender, or age.

The company behind Claude AI

Claude was developed by Anthropic, which bills itself as an AI safety and research firm. Based in San Francisco, Anthropic was founded in 2021 by former OpenAI (the company that makes ChatGPT and DALL-E) executives and researchers. Google and Amazon are major investors.

Claude release schedule

• Claude was first released in March 2023.
• Claude 2 was released in July 2023.
• Claude 3 was released in March 2024.

Claude AI vs. ChatGPT: Which is better?

It’s understandable to be curious about how ChatGPT and Claude AI compare because Claude was developed as a competitor to ChatGPT. Both chatbots have advantages and disadvantages. Therefore, it’s important to take certain factors into account when deciding which one to use.

How they use your data

For privacy-conscious users, it’s important to note that Claude and ChatGPT have different approaches to storing and using data.

Respect for privacy is one of the core pillars of Anthropic’s training processes. Anthropic doesn’t use your prompts or responses to train models unless you give them permission or the content is flagged for review. The company retains data on the backend for 90 days for individual users, though you can always see your prompts and responses within the tool.

OpenAI may use your conversations with ChatGPT for training unless you opt out. You can opt out by filling out a form or changing the settings on the mobile app.

How they’re moderated

Anthropic and OpenAI have measures to discourage toxic prompts and harmful responses. However, their approaches to content moderation are different.

Since Anthropic bills itself as an AI safety research company, it’s open and up front about its ethical practices. As part of Anthropic’s commitment to AI safety, all of its models incorporate the code of ethics outlined by Constitutional AI principles.

ChatGPT is fine-tuned for safety through a process called reinforcement learning from human feedback (RLHF). With RLHF, human reviewers rate the chatbot’s responses for bias, harm, and other unwanted characteristics.

According to the Anthropic team, their approach allows companies to scale oversight as AI models grow more sophisticated. Claude can self-moderate without the need for human resources and without exposing people to large amounts of toxic content. It’s also easier to observe how it performs against a set of principles and adjust when needed.

What they can do

Claude and ChatGPT offer different capabilities, depending on which version of the platforms you use.

Claude’s free tier is more expansive than ChatGPT’s. With the free version of Claude, you can upload files, which you can’t do with ChatGPT. Claude is also available via an app for Slack.

However, ChatGPT Plus can do more than Claude Pro. ChatGPT Plus offers image creation through DALL-E and voice chat, neither of which is offered with Claude Pro. ChatGPT is also available via a mobile app.

Knowledge cutoff

Most generative AI platforms have a cutoff date for their knowledge base, so they can only provide information up to a certain date. Claude has more up-to-date information than ChatGPT does. Claude’s training dataset ends in August 2023, while ChatGPT’s knowledge cutoff is September 2021. If you get ChatGPT Plus and use GPT-4, the information date cutoff is April 2023. However, ChatGPT is able to search the internet, so it can find up-to-date information as well. Claude cannot yet, but this may change since Anthropic just announced Tool use.

Plugins and extensions

If you’d like to integrate generative AI with other services, ChatGPT offers a marketplace of plugins that you can integrate with the chatbot. These add-ons can help you do things like search for travel accommodations or read webpages. Some plugins are built by OpenAI, while others are from services that you may already use, like Kayak and OpenTable.

Claude doesn’t offer any plugins.

Is Claude AI free to use?

Claude is available for free with daily usage limits. The limits vary based on demand.

The Claude Pro plan offers five times more usage for a monthly subscription. In addition to expanding the daily limits, Claude Pro offers priority access during periods of high demand, early access to new features, and the ability to use the latest, most intelligent model.

How to use Claude AI

You can use Claude AI for various purposes, from creating original content to talking through a tough decision. Here are some of the things you can try with Claude:

Engage in natural conversation

Claude is capable of understanding natural language prompts and dialoguing with people. You can use slang and idioms in your prompts, and Claude will understand them.

Here are some examples of conversations you can have with Claude:

• Bounce around party theme ideas.
• Engage in thought experiments on philosophical topics like Schrödinger’s Cat or the Ship of Theseus.
• Discuss which college major to choose by considering various factors.
• Walk through a complex math problem.
• Get recommendations for places to travel based on your interests and experiences.

Get educational or helpful information

You can use Claude to supplement research or ask questions you’re curious about.

For example, you can use Claude to do the following:

• Ask what tools you need to mount a television.
• Find out when the major bank holidays are in India.
• Walk through the steps for creating homemade pasta.
• Learn the names of influential European artists from the 17th century.
• Explore folklore from Central America.

Content writing

With Claude, you can write all kinds of text-based content. Simply provide the format of your content, what you want it to say, and the style you want to use.

Here are some ways to use Claude for content generation:

• Draft business memos.
• Produce an outline for a short story.
• Make a chore list for the family.
• Create prompts for journaling.
• Write a letter to your elected officials.

If you’re looking for an AI tool that can help you write more efficiently, consider Grammarly. Grammarly is an AI writing partner that can assist you in writing a variety of content, such as business documents, email replies, thesis papers, and more. Furthermore, unlike Claude AI or ChatGPT, you won’t have to worry about switching contexts. Grammarly works where you already write, making it a convenient and efficient tool for improving your writing.

Summarize large volumes of text

Claude can summarize text that you copy into the prompt box or upload in a file. It accepts several file types, including PDFs, TXTs, CSVs, and Docx files.

You can use Claude for the following tasks:

• Upload a PDF of a novel and ask for the main plot points.
• Paste a long email and ask for the tasks it outlines.
• Upload a report with charts and graphs and ask for a summary of the key data points.
• Paste the transcript of a speech and ask for the main themes.
• Upload meeting notes and get the most important takeaways for people in your department.

Coding

Claude can write code in most major programming languages, though it’s particularly effective with Python.

Try using Claude for these coding tasks:

• Insert a code fragment and get an explanation of what the code does.
• Take a snippet of old code and update it to be more efficient.
• Get feedback on best practices to use to improve existing code.
• Write new code to perform a specific function.
• Identify potential errors with existing code.

Visual processing

You can upload images to Claude and ask it to describe or analyze them.

Here are some ways to use this feature:

• Transcribe a handwritten note.
• Upload a photo of a plaque at a historical site and ask for more information about the location’s history.
• Upload an image of a place and use it as inspiration for a short story outline.
• Describe the items in an image for someone who is visually impaired.

The advantages of using Claude AI

Claude offers advantages that promote AI safety and allow you to use the chatbot efficiently, even for complex tasks.

Emphasis on safety

Meeting high ethical standards is a core part of Claude’s functionality. Anthropic developed Claude to demonstrate how to make AI safe. The company is very up front about its core views on AI safety:

• AI capabilities will rapidly accelerate and may one day equal or exceed human-level performance on many tasks.
• As AI grows more powerful, researchers must implement robust guardrails and ensure their platforms stay within them.
• By researching multiple approaches to AI safety, it’s possible to create helpful, honest, and harmless systems.

With Constitutional AI, Claude can train itself to avoid harmful requests and responses. Furthermore, the principles guiding Constitutional AI are publicly available. Generative AI platforms are advancing rapidly, and transparency around ethics benefits us all.

Large context window

All generative AI platforms have thresholds for how much information they can process before the platform starts to lose the context of that data. This is called the context window. Claude’s context window is particularly large at 200,000 tokens, which is equivalent to 350 pages of text.

This large context window makes Claude particularly powerful for parsing through lengthy documents. You can also add more context to your prompts. For example, if you’re brainstorming marketing ideas, you can upload a detailed research report on your audience, market, and previous performance. Claude can use that report to create a more tailored, relevant response.

Having a larger context window also allows for more extensive, relevant responses. Claude can generate long-form content and stay on topic, which is particularly useful for complex, nuanced subjects.

Speed

Claude is highly efficient. According to Anthropic, Claude 3 can process roughly 30 pages of text per second and read dense research papers in less than three seconds, three times faster than its peers. This speed, and its lengthy context window, further enhance Claude’s ability to process large documents quickly.

The drawbacks of using Claude AI

No generative AI platform is perfect. Claude has some drawbacks related to its capabilities and potential inaccuracies.

Limited capabilities

Because Claude is a text generation tool, its capabilities aren’t as expansive as some other platforms. If you are looking for a generative AI platform that can create text and images all in one tool, you can’t do that with Claude.

It also doesn’t offer plugins, which allow you to customize your experience with generative AI. For instance, if you’re on a fitness journey, you can use plugins that calculate calories and macros for various meals. Because Claude doesn’t offer plugins, you can’t incorporate those extra customizations.

Lack of up-to-date information

Claude’s knowledge base is updated regularly, but it’s typically several months old. It doesn’t have real-time information and cannot search the internet. If you want up-to-date information on topics like the economy, laws and regulations, and pop culture news, Claude may not be your best option.

Inaccurate responses

Although Claude was trained on a massive amount of data, it can sometimes generate inaccurate responses. These inaccuracies, called hallucinations, are an issue with all generative AI platforms. Generative AI was made to create content that could have been convincingly made by a human, not to verify that content for accuracy.

Because of these hallucinations, it’s important to verify Claude’s responses, especially in cases where accuracy is critical, such as in academia or business communications.

Claude AI: A chatbot built for safety

Claude is a generative AI chatbot built with ethics and safety in mind. Because it incorporates the Constitutional AI model, Claude follows a set of principles to generate helpful, honest, and harmless responses.

Claude’s code of ethics, speed, and ability to process large volumes of information enable you to efficiently leverage AI for complex analysis and content generation. However, it’s important to be mindful of potential inaccuracies and limited capabilities.

Claude is one of the many generative AI tools disrupting how we create, research, and interact with the systems we use. These tools are constantly changing and evolving. Keeping an open, curious mind will allow you to keep up with the pace of change and use generative AI safely and wisely.

The post Claude AI 101: What It Is and How It Works appeared first on Grammarly Blog.

GPT-4 is a versatile generative AI system that can both interpret and produce a wide range of content. Learn what it is, how it works, and how you can use it to create content, analyze data, and much more.

Table of contents

• What is GPT-4?
• Who created GPT-4?
• How GPT-4 works
• Is GPT-4 free?
• GPT-4 capabilities
• GPT-4 API use cases
• Advantages of GPT-4
• Limitations of GPT-4
• Conclusion

What is GPT-4?

GPT-4 is a highly adaptable generative AI tool that supports multimodal inputs. This means it is capable of interpreting and processing a wide range of content, not just text, but also audio and images. Users can feed it various types of data, and in return, GPT-4 can produce outputs that include detailed written passages, in-depth explanations, computer code, and creative compositions, all in a manner that closely mimics human thought and language patterns.

What makes GPT-4 different from ChatGPT

GPT-4 and ChatGPT are closely related but not the same. ChatGPT is a chatbot with which people can have conversations with the underlying Large Language Model (LLM). Essentially, ChatGPT is the conversational interface to the model. You can enter text prompts in natural language, and ChatGPT will respond with answers to your prompts.

ChatGPT can run on various versions of the GPT model. By default, the free version of ChatGPT gives you access to GPT 3.5. With a paid subscription, you can get access to GPT-4.

GPT-4 vs. GPT-4 Turbo: What’s the difference?

GPT-4 Turbo is a faster and more cost-effective version of GPT-4 that’s suitable for large-scale applications. In fact, the most recent version of GPT-4 Turbo is more affordable and capable than GPT-4. GPT-4 Turbo also has a longer context window, which means you can send up to 300 pages of text in your input prompts.

Overall, the choice between GPT-4 and GPT-4 Turbo depends on an application’s specific requirements, particularly in terms of response complexity, speed, and operational costs.

Who created GPT-4?

OpenAI, an artificial intelligence firm in San Francisco, created GPT-4. OpenAI was founded in 2015 to create artificial intelligence that’s “safe and benefits all humanity.” The company is behind several leading AI platforms, including DALL-E and Codex.

OpenAI released GPT-4 on March 14, 2023.

How does GPT-4 work?

GPT-4 doesn’t pull its responses from a database of knowledge. It generates one word at a time, predicting each word as it goes. Its predictions are based on statistical patterns it identified by analyzing large volumes of data.

The technology that makes this advanced analysis possible is called a Generative Pretrained Transformer (GPT). GPT is the name given to a family of LLMs made by OpenAI. Let’s look at how researchers train GPT models to better understand how they work.

How GPT models are trained

The GPT model training process is broken up into two stages: pre-training and fine-tuning.

During pre-training, the model processes and analyzes large volumes of data from the internet and licensed data from third-party sources. It identifies patterns and correlations between words and images to understand meaning and context. It also learns the structures of sentences, paragraphs, and various types of content, like poetry, academic papers, and code.

GPT models use an advanced neural network architecture called a transformer. The transformer is key to the model’s ability to parse through large volumes of data and learn independently. The transformer allows the model to process and learn patterns from the training data, which enables GPT models like GPT-4 to make predictions on new data inputs.

The next stage of training is fine-tuning. At this stage, the model is refined to perform specific tasks, such as generating conversational responses. The model learns how to provide the answers people want through reinforcement learning from human feedback (RLHF). Humans rate the model’s responses, and the model tries to get more positive feedback with each subsequent response. The fine-tuning stage is also an opportunity to minimize biases and reduce harmful responses.

Previous GPT models

GPT-4 is the fourth iteration of OpenAI’s GPT models. Here’s an overview of how the model family has evolved.

• GPT-1 was introduced in 2018. It was trained on BookCorpus, which consists of 7,000 unpublished fiction books. This model proved that the GPT framework could achieve a natural language understanding.
• GPT-2 was introduced in February 2019. It was trained on 8 million webpages. The training goal was to create a model to predict the next word in a text after being given all the previous words. Researchers pushed the model beyond its training by asking it to generate arguments. The result was an essay that a human could have written. Although GPT-2 performed inconsistently, it could answer questions, translate text, and summarize long content.
• GPT-3 was announced in the summer of 2020. OpenAI referred to it as a general-purpose text generation platform. The dataset that trained GPT-3 contained more than one trillion words. Unlike its predecessors, GPT-3 could generate code. GPT-3 acted as the base for ChatGPT, the AI-powered chatbot.

GPT-4 training and key capabilities

OpenAI began creating the deep learning tools used to build GPT-4 in 2021. It worked with Microsoft Azure to develop a supercomputer capable of handling the computing power and volume of data that advanced LLMs require.

GPT-4 was trained on publicly available data and data from third-party sources. Unlike previous models, OpenAI hasn’t released any information about the size of the training model, the hardware it used, or details on the training methodology.

What we do know is that GPT-4 is more advanced than GPT-3 in several ways:

• Can accept both images and text-based prompts
• Was trained on data up to April 2023; GPT-3’s dataset stops at June 2021
• Performs better at creative tasks than GPT-3
• Able to handle more complex tasks than its predecessor, such as analyzing graphs
• Can handle longer prompts up to 25,000 words
• Is more likely to stay within guardrails for allowed content
• Generates more accurate responses
• Is better at adapting to user requests, such as your brand personality or writing style

OpenAI also used several tests to validate GPT-4’s aptitude. It performed well on AP exams, the Uniform Bar Exam, the Olympiad Exam, the LSAT, and the GRE Quantitative exam.

Is GPT-4 free?

You have to pay to use GPT-4 directly from OpenAI. There are two ways to access it.

With a paid subscription to ChatGPT Plus, you get access to GPT-4. You can then converse with ChatGPT on the web or with apps for Android and iOS.

Developers can access GPT-4 through the Developer API. With the API, you pay a set rate for tokens. There’s one rate for prompt tokens—the tokens you use in your “question” to the LLM, and another for completion tokens, the tokens used in the “answer” you receive from the LLM.

Here’s how tokens work:

• Each input and output is broken down into tokens. Prompt tokens refer to the text and files you provide in your request to GPT-4. Completion tokens refer to the text generated by GPT-4 in its response.
• Before GPT-4 processes your request, the input is broken down into tokens. These tokens are not the same as syllables, or logical word fragments, they can include spaces or sub-words.
• There are a few rules of thumb to understand the “exchange rate” between words and tokens. In English, four characters roughly translate to one token, and seventy-five words roughly translate to 100 tokens. In other languages, this ratio does not hold, and each word likely translates to a higher number of tokens.

Another way to access GPT-4 is through Microsoft’s Copilot AI. Copilot is a chatbot that runs on GPT-4. Copilot is available online and through mobile apps.

What you can do with GPT-4

GPT-4’s ability to interpret nuance, process more complex prompts, and accept images means it has a wide range of potential applications. However, like all current AI systems, GPT-4 has limitations that require thoughtful use.

Let’s start with some ways you can use it within the ChatGPT platform.

Analyze images

You can upload an image in GPT-4 and ask to perform tasks based on that image. Here are some of the image analysis tasks you can request GPT-4 to complete:

• Interpret data in a chart or graph
• Describe an image, including what the subjects of the image are doing and how many of them there are
• Read and analyze photos of text, such as historical documents
• Turn handwritten notes into text
• Identify what’s funny, sad, or surprising about an image

Generate text

GPT-4 can generate original text content for formal communications, business activities, or personal tasks. Here are a few examples:

• Write training materials
• Create procedural documents, handbooks, and policies
• Translate content in different languages
• Answer basic research questions, like how many provinces are in Kenya or how air purifiers work

Generative AI is widely used for text creation, but if you need a writing tool that integrates seamlessly with your current workflow, Grammarly might be the better choice. It’s employed by individuals and teams alike for brainstorming, composing, and revising content directly within over 500,000 apps and websites. This eliminates the need to copy and paste your work between platforms.

Generate creative content

GPT-4 boasts better creative writing capabilities than its predecessor, GPT-3.5. In particular, it’s better at maintaining the cohesiveness and consistency of a narrative.

Here are some ways to use these capabilities:

• Create fictional creatures with descriptions of how they look, their history, and lore
• Describe an image with prose written in a particular style
• Outline a short story
• Draft blogs, social media captions, and marketing communications content
• Explain a complex topic, like software development, in the format of a poem

Write code

GPT-4 can write, translate, and optimize code in dozens of programming languages. You can generate and analyze code in several ways:

• Upload a drawing of a website layout and ask GPT-4 to generate code that matches it
• Describe what you want the code to do in natural language
• Paste in existing code and ask GPT-4 to identify errors
• Get an easy-to-understand description of what a snippet of code does

Summarize and analyze content

GPT-4 can parse through large volumes of data to track data trends, summarize texts, and explain content. You can enter text directly into the application or upload files in every popular format.

GPT-4 can read and analyze content for a variety of applications:

• Identify sales trends in an Excel document
• Write a 250-word summary of a long, complex text, like an academic article
• Find similarities between two articles
• Explain the plot of a short story, with details about the writing style and themes
• Review texts and provide suggestions for improvement

GPT-4 API use cases

Developers use the GPT-4 API to create new applications and add features to existing ones. Here are some of the more common categories these applications fall into.

Content generation

Although ChatGPT can generate content with GPT-4, developers can create custom content generation tools with interfaces and additional features tailored to specific users. For example, GPT-4 can be fine-tuned with information like advertisements, website copy, direct mail, and email campaigns to create an app for writing marketing content. The app interface may allow you to enter keywords, brand voice and tone, and audience segments and automatically incorporate that information into your prompts.

Chatbots

GPT-4 can serve as the basis for conversational AI platforms. Developers can create custom chatbots for specific functions, like customer service, embodying a character or historical figure, or answering homework questions.

Custom assistants

GPT-4 can power AI assistants tailored to specific industries, professions, or interests. For example, you can create an assistant for legal professionals or for brainstorming creative ideas.

Sentiment analysis

GPT-4 can serve as the basis for sentiment analysis apps, which scan reviews and social media to find common themes in customer feedback and public opinion.

Assistive technology

GPT-4 opens up new possibilities for making the world more accessible. For example, it can provide text descriptions of images for visually impaired people.

Advantages of GPT-4

GPT-4 offers many features and functionalities. Here are a few examples of GPT-4’s capabilities.

It’s multimodal

GPT-4’s ability to accept images, files, and text enables it to perform complex tasks. These multimodal capabilities expand the potential of nearly every GPT-4-based application.

Here’s how you can benefit from GPT-4’s multimodality:

• Add greater context and depth to prompts using multiple sources. For example, a restaurant chain can use GPT-4 to scan photos and captions from social media to assess customer sentiment. This allows them to do more than capture positive and negative words in social posts. They can also see which photos of food items tend to have positive captions and which ones tend to have negative captions.
• Save time. Since you can add attachments directly to the platform, you don’t have to write your own summary of the file or image related to your prompt. GPT-4 can also automate tasks like product descriptions and reports. Simply upload an image or raw data and prompt GPT-4 to generate a response that fits within your guidelines.
• Create multi-step prompts. GPT-4 can take information from an image and perform complex tasks with it. For instance, you can upload a photo of a rehearsal schedule for a play and ask GPT-4 which days and times the lead characters are scheduled to rehearse.

It’s better at understanding nuance

GPT-4 is especially good at detecting nuances like emotion, dialects, and colloquialisms in written text. It can also infer meaning without you having to say things directly.

The ability to understand nuance makes GPT-4’s output even more human-like:

• Generate authentic-sounding dialogue between characters from different places
• Assess the emotions of people in an image and write content targeted to those emotions
• Allow humans to write natural-sounding prompts and respond with contextually accurate content

It’s flexible

Although chatbots are some of the most popular applications created with GPT-4, the model can power many generative AI applications. This is because you can fine-tune GPT-4 on your own dataset. Then, you can integrate it with existing applications or create new ones that look and feel like your brand. Because of that flexibility, developers in every field, from medicine to consumer goods, can innovate with GPT-4.

Here are some of the ways you can use GPT-4’s flexibility:

• Offer customers self-service tools
• Enable non-technical people to do technical tasks, like coding
• Create custom recommendations for music, books, podcasts, etc.
• Automate manual tasks, like medical documentation

Disadvantages of GPT-4

GPT-4 is an advanced generative AI platform, but it has drawbacks. Here’s what to be on the lookout for when you use it.

It can produce inaccurate answers

All generative AI platforms are prone to producing inaccurate information. AI researchers refer to these inaccuracies as hallucinations. Although GPT-4 is more accurate than its predecessors, it doesn’t verify information and it doesn’t know when it’s wrong. Its creators mention that it can be confidently wrong. Because of these inaccuracies, developers should be thoughtful when considering whether to integrate GPT-4 into their applications. If the application has limited error tolerance, then it might be worth verifying or cross-checking the information produced by GPT-4.

It has a limited information base

GPT-4’s training dataset only goes up to April 2023, which means that it doesn’t include the latest news and trends in its responses. If you use GPT-4 for research, it won’t have up-to-the-minute insights. It may be out-of-date on topics like technology, where information changes quickly.

Developers can work around this limitation by fine-tuning the model with more up-to-date data or creating applications that add online search capabilities to the model.

It can be costly to access

The only way to access GPT-4 for free is through Microsoft’s Copilot AI. If you prefer to use it through ChatGPT, it costs at least $20 per month. Depending on your needs and your budget, that may be prohibitive.

Furthermore, developers might find the API access to GPT-4 to be expensive, especially if they are running a popular application that uses a lot of tokens.

GPT-4 and the generative AI landscape

GPT-4 is one of the leading generative AI platforms because of its advanced processing abilities, multimodal capabilities, and flexibility. Everyday users can create original content with GPT-4 through a premium subscription to ChatGPT. Developers can use the API to build new applications and improve existing ones.

Though GPT-4 has many applications, its inaccuracies and costs may be prohibitive for some users. However, it’s just one of many generative AI platforms. Keep your ear to the ground to stay updated on the latest AI tools and what you can do with them.

The post GPT-4 Basics: How It Works and How to Use It appeared first on Grammarly Blog.

It’s been said that artificial intelligence is not going to eliminate everyone’s job; it’s going to eliminate jobs for people who don’t know how to use AI. This sentiment may be especially applicable to brand marketing. Even in its infancy, generative artificial intelligence, or gen AI, is having a revolutionary impact on brand storytelling. With the help of gen AI, hyper-personalized brand narratives and immersive experiences are no longer a marketer’s dream; they’re today’s reality, and knowing how to create them is essential for marketers everywhere.

Read on to learn how to create an irresistibly compelling, intricately tailored, and deeply impactful brand story with AI.

What is brand storytelling with AI?

Brand storytelling with AI is the use of gen AI’s data-harnessing creativity to generate personalized, data-driven brand narratives that are deeply engaging and impactful.

To understand how to maximize the bold new marketing frontier of AI-driven branding, it’s important to understand the basics of brand storytelling and AI.

What is brand storytelling?

Brand storytelling is the art of connecting with your audience through narratives that convey your brand’s purpose, values, and personality to evoke empathy, alignment, and a deeper connection. It’s an important part of earned media, paid media, social media, print advertising, editorial, and many other brand communications.

Far from simply telling an audience brand facts, brand storytelling is an immersive, attention-grabbing experience that fosters brand loyalty, sets you apart from the competition, and helps your audience establish a relationship with the brand.

A compelling brand story allows the audience to imagine themselves at the center of the story. For example, it can feature a story about connection or a sense of belonging, highlight an experience of overcoming adversity or growth, or present an enthralling personal narrative. In all cases, the backbone of great brand storytelling is a consistent, cohesive brand message that echoes your brand identity, evokes an emotional response—and often action—and aligns you with your audience through shared motivations, interests, and values.

What is gen AI?

AI, short for artificial intelligence, is a broad term used to describe machines programmed to learn and perform tasks that normally require human intelligence, like problem-solving, reasoning, and decision-making. Generative AI, or gen AI, is a type of artificial intelligence that learns from existing data and then uses its learnings to generate text, images, code, music, and video.

The intersection of gen AI and brand storytelling

Brands have been using images, words, video, audio, and interactive elements to help tell engaging brand stories long before there were significant ways to measure results. Advancements in data analysis and technology allow brands to better understand who their customers are and how to speak to them. But there has still been a significant amount of guessing about what customers want and what engages them most.

Gen AI changes all that, marking a paradigm shift in brand narrative development. Now, creative AI solutions and digital storytelling techniques can leverage data and machine learning to produce a wildly immersive, high-quality, and impactful brand story with AI in seconds.

Understanding gen AI and its capabilities

Though still at the beginning of its evolution, gen AI has the potential to help you make exceptional content with minimal budget and maximum returns, all based on data and a deeper understanding of your audience. This extends even to using AI in advertising.

The more gen AI learns about your brand through existing data and your teachings, the more it can act as a knowledgeable co-creator, helping you to weave personalized, on-brand stories with rich visuals, text, and audio based on audience emotions and even predictions of your audience’s upcoming desires. In other words, gen AI is positioned to become one of the most powerful tools a marketer can use to create impactful, relevant content that thrills the target audience and puts them right at the center of every story.

Customer alignment isn’t the only benefit of creating a brand story with AI. Historically, producing the highest-quality content required a lot of planning, talent, time, money, and effort. Gen AI will make high-quality, on-brand content more accessible to anyone with access to a gen AI platform and the skills to use it; with AI, content that was once impossible to produce alone or required significant, costly production—think a production team, cameras, lighting, sets, designers, photographers, recipe developers, ingredients, translators, writers, support staff, you name it—can be conceptualized and executed from your computer.

5 steps to create a brand story with AI

There are an increasing number of platforms that help you create a brand story with AI, from text to images and beyond (think ChatGPT, Gemini, Cohere, DALL-E 2, Midjourney, StoryLab.ai, Ink Lantern, Jasper, Synthesia, and beyond). Regardless of which gen AI platforms you use, follow these steps to ensure you get the most out of AI-enhanced brand stories:

1 Define brand values and objectives.

Your brand values and objectives are your company’s compass. Identifying them and integrating them into your gen AI prompts, when possible, has several benefits; it mitigates the potential for AI misdirection, ensures your brand story echoes your brand identity and voice and resonates with your audience, and enables you to measure and achieve your marketing goals.

2 Use gen AI for audience analysis and insights.

Gen AI isn’t just a creative resource. It’s a powerful tool for understanding your audience’s behavior, thoughts, and emotions. For example, you can use it to analyze text in social media mentions, forum discussions, and customer reviews and feedback—including those in other languages—to get insight on your customers’ perceptions of your brand, identify influencers your customers trust, and reveal other important information from unstructured text data.

AI can also categorize customer feedback and identify recurring issues, resulting in valuable insight into customer satisfaction and areas that need improvement. It can even predict customer churn, allowing you the opportunity to consider new retention strategies.

Gen AI analysis and insights extend to image and video analysis. This is profoundly helpful because it can help you understand and leverage the visual preferences of your target audience, identify brand ambassadors interacting with your brand (gen AI can recognize logos), detect trends, and direct your visual content strategy to focus on what resonates and has the most emotional impact.

Finally, gen AI can utilize its learnings to predict users’ future interests and behaviors, allowing you the opportunity to personalize content, messages, and recommendations for the greatest impact.

3 Develop a narrative framework using AI-generated ideas.

During brand-narrative development, a strong, clear framework ensures a cohesive, engaging narrative that resonates with your audience on a deeper level. With your brand’s values, objectives, and analytics in place, you can design the blueprint for your brand’s story or prompt gen AI to help you formulate it. Include key characteristics, plot points, and emotional arcs of the story in your prompts, then further refine the ideas to fit your brand.

4 Prompt AI to create or refine your story.

You may have clear ideas of how you want to tell your story, but don’t overlook the opportunity for AI to help you craft it or explore unexpected creative angles. Whether you need words, images, audio, or all three, use your established story framework as a prompt in your gen AI platform of choice to create, polish, or refine your brand story. To get the best results, provide clear, detailed prompts in your brand’s voice, then prompt AI to refine as needed. Once you’re satisfied with the results, expect to do a little editing, rewriting, or personalizing in your brand’s voice; gen AI is a helpful tool, not a replacement for your creativity and brand knowledge.

5 Personalize and refine the story for different platforms and audiences.

One of the beauties of AI is you can prompt it to iterate on its creations as well as your own creations. This makes it infinitely easier to further customize your story for various digital platforms, taking performance and audience data into account to garner the most impactful results.

Refining your brand narrative with Grammarly’s brand tones and other gen AI capabilities

Whether you’re crafting words to describe the framework of a brand story using AI or are using them in the stories themselves, it’s important to write clearly and professionally using your brand voice. Grammarly, an AI-powered writing assistant, can help you define your brand voice and ensure your copy follows suit with crisp, grammatically correct copy and proper punctuation.

Grammarly is the web’s go-to writing assistant for comprehensive feedback on spelling, grammar, punctuation, and clarity to ensure mistake-free writing, and it’s expanding its suite of AI-powered writing support. For example, its brand tones feature allows you to conceptualize, set, and save brand tone profiles. Once your tones are set up, you can opt for Grammarly’s feedback and recommendations to be customized to fit your brand’s unique voice for all of your various communications. You can also continue to refine your brand tones so Grammarly’s feedback is always tailored to your brand voice.

Grammarly’s generative AI capabilities also help you draft contextually relevant content for blogs, replies, and more—on-demand, no matter what you’re working on in your browser.

Equally important as we all begin to rely more on AI writing, Grammarly can detect plagiarism, allowing you to ensure all of your content is original.

Key takeaways on creating a brand story with AI

Though generative AI is in its early stages, it’s already revolutionizing brand storytelling by using machine learning to produce highly targeted and personalized, immersive, and emotion- and action-evoking brand stories.

To stay competitive and truly connect with your audience, marketing managers everywhere will benefit from embracing the unprecedented marketing innovations associated with co-creating with gen AI. Beyond enhancing your brand’s narrative and connecting with your audience in profound new ways, you’ll grow alongside this new marketing technology and be empowered to use it to its fullest potential with each new advancement.

Brand storytelling with AI FAQs

What is brand storytelling with gen AI?

Brand storytelling with AI is the act of using gen AI’s data-harnessing capabilities to generate personalized, data-driven, brand narratives that are deeply engaging and impactful.

Can gen AI replace the need for human creativity in branding?

Gen AI is not a replacement for human creativity in branding. It’s a co-creator that allows marketing managers and teams to create more immersive, hyper-personalized stories by tapping into AI’s data-driven knowledge about your brand and its audience.

How can a marketing team with limited AI expertise use gen AI?

Starting with any gen AI platform is an important step toward literacy. To get started, explore Grammarly’s brand tones to help refine your brand voice and streamline your communications, then select a tool that suits your needs; perhaps brainstorm for new content ideas with ChatGPT or Gemini or create images with Midjourney or DALL-E 2.

The post Brand Narratives: A Guide to Creating a Brand Story With Gen AI appeared first on Grammarly Blog.

Research papers rely on other people’s writing as a foundation to create new ideas, but you can’t just use someone else’s words. That’s why paraphrasing is an essential writing technique for academic writing.

Paraphrasing rewrites another person’s ideas, evidence, or opinions in your own words. With proper attribution, paraphrasing helps you expand on another’s work and back up your own ideas with information from other sources while retaining your own writing style and tone.

In this guide to paraphrasing, we explain how to strengthen your research papers through the art and craft of paraphrasing. We discuss the rules of ethical paraphrasing and share paraphrasing tips to help you get started. We even provide a few paraphrasing examples to illustrate how to do it yourself.

Why should you paraphrase in a research paper?

There are a few reasons research writers rely on paraphrasing in their papers:

• It shows comprehension. Paraphrasing requires you to understand ideas well enough to write them in your own words, so it not only helps you pass on information but also can help you learn and retain it.
• Paraphrasing other research or another writer’s work allows you to make valuable connections between ideas. Crediting your sources ethically and according to standards shows professional collaboration and respect.
• Paraphrasing can transform dense academic language into clearer or more modern text. Research writers employ it to make important information more understandable to a wider audience.
• Paraphrasing can increase the readability of your paper and make impactful direct quotes stand out.

When should paraphrasing be used in a research paper?

Paraphrasing is best used in concert with other research writing techniques, such as direct quotes and summaries. Here are instances when paraphrasing is appropriate for your research paper:

• Opt for paraphrasing when you can explain the same concept in plainer language or with less jargon.
• Paraphrasing works best when you need to share background information. Save direct quotes for striking statements and opinions. Rely on your own words to set the stage or provide context.
• Similarly, methodology from published studies generally doesn’t require direct quotes. Consider rewriting this contextual information in your own words.
• Paraphrasing also works well when you’re reporting key results from other research. You might restate the results by paraphrasing the main findings and then use a direct quote to share opinions about the value gleaned from the research.

Paraphrasing vs. quoting and summarizing

Unlike summarizing, paraphrasing uses roughly the same amount of detail as the original work but adjusts the language to demonstrate comprehension or make the text more understandable. Summarizing, in contrast, shortens the information to only the most important points.

While paraphrasing uses your own phrasing, quoting transcribes someone else’s words exactly, placing them in quotation marks so the reader knows someone else said them.

Direct quotes work best when you’re dealing with striking statements or opinions or when you want the tone of the original work to shine. Opt for paraphrasing when you can convey the same information in plain language. Sometimes, placing a direct quote in a sentence would lead to an error in subject-verb agreement or pronoun agreement, so paraphrasing works better in that case. Paraphrasing can also help modernize outdated wording, such as gendered language.

Generally, your writing will have the most readability and engagement if you strike a balance between paraphrasing and direct quotes.

Common paraphrasing mistakes

Writers risk committing plagiarism or losing clarity when they commit the following common paraphrasing mistakes:

• Substituting synonyms but not otherwise changing the phrasing
• Altering the original meaning
• Failing to add citations within the text and in the bibliography

Tips for paraphrasing successfully in your research paper

Try to rewrite from memory

It can be difficult to reword a passage when you’re staring at it. Sometimes it can help to jot down notes about a passage and then try to rewrite the same sentiment from scratch. This forces your brain to think creatively because you can’t just copy the passage verbatim.

Focus on meaning, not just vocabulary

Paraphrasing is more than just swapping out words for their synonyms; you need to completely rewrite a sentence in your own style. Pay close attention to what the original author is trying to say as a whole, rather than focusing on the individual words. You may find yourself changing phrases or clauses. You may even come up with a way to restate the whole idea in a clearer or more concise way.

Change or update the language

Use synonyms to replace the essential words of an original passage with other words that mean the same thing, such as using “scientist” for “researcher,” or “seniors” for “the elderly.” You can also pay special attention to modernizing and broadening the language, such as for more gender inclusivity. This is a common approach to paraphrasing, although it’s not sufficient on its own.

Edit the sentence structure

Editing the sentence structure by rearranging the order of certain phrases and clauses or combining or breaking apart sentences is another strategy for paraphrasing. But if you do this, be careful not to overuse the passive voice.

Sometimes, you can rephrase a sentence by changing the parts of speech, such as converting a gerund into an operative verb or turning an adjective into an adverb. This strategy depends on the wording of the original passage, so you may not always have the opportunity.

Often, using only one of these techniques is not enough to differentiate your paraphrase from the source material. Try combining a few of these techniques on the same passage to set it apart.

Use transition phrasing

Some introductory and transitional phrases let your reader know you’re about to paraphrase an existing work. This tactic has the added benefit of helping you rewrite key findings by recasting the sentence structure with a new subject. Here are a few examples:

• Research shows that . . .
• A recent study found that . . .
• According to [author]’s analysis . . .
• Thanks to [source], we now know that . . .

Avoid patchwriting

If you don’t change enough of the original, it leaves “patches” of the source text that are easily identifiable to anyone who’s read it. This is known as patchwriting, and it’s a big problem with paraphrasing. Double-check to see if your paraphrase is unique enough with our free plagiarism checker.

Use ethical paraphrasing tools

Use Grammarly’s free paraphrasing tool to quickly paraphrase text with the help of generative AI. Paste the text into Grammarly to get options for how to paraphrase it instantly, then use our citations generator to get the proper attribution.

Learn about other aspects of research paper writing by browsing Grammarly’s research paper guides and resources.

Paraphrasing examples

Paraphrasing a research paper to avoid plagiarism

Plagiarism refers to claiming another person’s ideas or words as your own. Paraphrasing alone is not enough to avoid plagiarism—if the words are different but the ideas are the same, you have to do more. That’s why citing paraphrases is not just morally right, it’s also a mandatory part of how to write a research paper, regardless of the research paper topic.

In academic writing, paraphrases typically use parenthetical citations, a type of in-text citation that places the author’s last name in parentheses, along with the year of publication or page number. Parenthetical citations are placed at the end of a passage, before the ending punctuation.

Additionally, you need to include a full citation for any source you use in the bibliography section at the end of the research paper. A full citation includes all the necessary details the reader needs to track down the source, such as the full title, the publication year, and the name of the publisher.

The information to include in both parenthetical and full citations depends on which formatting style you’re using: APA, MLA, or Chicago. Refer to our guides to learn more about how to properly cite your paraphrasing in whatever style you prefer.

If you’re still having trouble citing paraphrases, you can use our free citation generator to save time.

How to paraphrase for a research paper FAQs

When should you use paraphrasing in research writing?

If you want to use someone else’s ideas in your research paper, you can either paraphrase or quote them. Paraphrasing works best when the original wording has room for improvement or doesn’t fit in with the rest of your paper. Quoting is best when the original wording is already perfect.

What techniques can you use for paraphrasing practice?

The most common paraphrasing technique is using synonyms to replace some of the original words. That only gets you so far, though; also consider rearranging the sentence structure, adding/removing parts of the original, or changing some of the parts of speech (like turning a verb into a noun).

Do research paper paraphrasing rules change for different citation styles?

The rules for paraphrasing are always the same—but the rules for citations change a lot between styles. Review the citation guidelines for the formatting style you’re using, whether APA, MLA, or Chicago.

Can I paraphrase sources with no named author, like websites?

Yes, you can paraphrase websites, but ensure they are reputable. And you still need to cite the source according to the citation guidelines.

What’s the best way to integrate paraphrased information smoothly in my paper’s flow?

Transitional phrases can help you introduce paraphrased information. Try using language such as:

• Research shows that . . .
• A recent study found that . . .
• According to [author]’s analysis . . .
• Thanks to [source], we now know that . . .

Use paraphrasing alongside other writing devices, such as direct quotes or summaries, to help your paper flow naturally.

Is it acceptable to paraphrase content from my own previous papers?

Yes, you can paraphrase your other content, unless your academic institution has a policy against it. You should still cite the original source material, even though it is your own work.

The post Paraphrasing for Better Research Papers: A Step-by-Step Guide appeared first on Grammarly Blog.