All popular features of awesome AI products
Explore over 20+ features built into many of the best AI products on the market as of January 2024.
This article is a high-level guide to the main features of the awesome AI products that are emerging as of January 2024.
I hope this inspires you to build the next great AI unicorn, or the killer feature that will unlock the next $100 million in revenue. If you want to support us and read new articles about AI products like this one, subscribe to the newsletter on Substack.
Table of content
-
Interfaces with AI
- Chat interface
- Playground interface
- Playground integrated into a workspace
- Mix chat & playground
-
Interactivity features
- Multimodality
- Auto-completion suggestions
- Auto-corrections
- Highlighted potential improvements
- Regenerate
- Multi-output
- Follow-up questions
- Copy, download, or share the output
- Interactive chain of thoughts
- Multi-language support
- Translation
- Agent (or GPT), and model selection
-
Personalization features
- Change of tone
- Change of style
- Mimicking a role, character, or person
-
Explainability features
- References to data sources, tools, and agents used
- Highlighted changes
- Displaying the chain of thoughts
1. Interfaces with AI
Chat interface
This interface involves a chat between a user and an AI, with the AI responding to each message from the user. It is inspired by human-to-human interaction, and in most cases, the user takes the role of the master, while the AI takes the role of the assistant, and does everything the master asks or entertains the master.
Playground interface
In this context, a "playground" refers to a space where users can tweak and experiment with AI-generated results until they achieve their desired outcome. Unlike the traditional question-and-answer structure of AI chat, playgrounds provide a rich environment that encourages brainstorming and creative expression. Take writing assistants like DeepL Write, for instance. It not only suggests an accurate new version of your content, but also suggests ways to restructure your sentences, replace words with better terms, change your tone, and adjust your writing style, providing users with a rich spectrum for refining and improving their written work.
Mix chat & playground
GitHub Copilot AI is a great example of chat and playground integration into the workspace.
2. Interactivity features
Multimodality
Enrich the user experience by combining text with other media elements like images, videos, or interactive elements.
Generative AI products can support multiple formats as inputs, outputs, or both. For example, you can ask Microsoft Bing what animal is in a picture by mixing text and an image in your input query, and Bing will also answer with a mix of text and images in its output.
Auto-completion suggestions
GitHub Copilot is a fantastic example of generative AI auto-completion suggestions. As programmers write code in their IDE - their integrated development environment - GitHub Copilot makes suggestions about what code to write next.
The suggestion is a passive hint, often displayed in gray, that programmers can ignore and program without being slowed down. If programmers find the suggestion useful, they can accept it, or accept the first word of the proposed.
With this core feature, GitHub Copilot was an instant success, adopted by millions of programmers and reducing coding time by approximately 50% according to GitHub surveys.
Auto-corrections
Some AI assistants automatically correct your input as you write. DeepL Write, for instance, corrects and improves your writing while highlighting in green what it has modified.
Highlighted potential improvements
As an alternative to auto-corrections or autocomplete improvements, Grammarly highlights what could be changed and provides a color to indicate why it's suggesting a change to improve the writing. For example, it highlights the grammatically incorrect text in red and the writing style improvements in yellow. The user can then click on the underlined word and Grammarly will display the AI-generated suggested corrections.
Select text & modify
To seamlessly integrate a generative AI playground into your workspace, both Notion AI GitHub Copilot and Microsoft Copilot allow you to select some text and trigger the AI assistant on the selected text. This way the AI knows where to focus, while also knowing that the text before and after the selected text might be useful as context.
Typically, a small AI playground is displayed on a layer above the workspace, where users can interact with the AI until it achieves its goal, which may be to fix grammar mistakes, change style, or find coding errors in the selected text.
Regenerate
Since the AI will always give the user what they want on the first try, many AI platforms allow the user to click the "Regenerate" button and the AI will try again.
Multi-output
An alternative to the "Regenerate" button is to provide multiple outputs and let the user decide which one is best. This is often the default option for image-generating models such as DALLE-3 and Midjourney, while it is often a secondary, more hidden feature in languages such as Bard or ChatGPT.
A great example of multi-output integration within a language model-based product is DeepL Write, where the user can decide among different options to restructure an entire sentence or replace a word.
Follow up questions
Follow-up questions in a chatbot prompt the user for more information or clarification, enhancing contextual understanding and engagement. They guide users, provide personalized responses, and contribute to a more interactive and user-friendly experience.
References to data sources
Adding links to the reference source where the AI assistant found the information allows the user to extend the AI output by reading the related data source.
For example, you can ask Microsoft Copilot for information about your colleagues' work. Microsoft Copilot will provide you with a short answer to your question, as well as a link to the documents where the information was found, which you may also want to read in full.
Copy, download, or share the output
AI playgrounds and chats often allow the user to copy or download the output generated by the AI for further use or to share with colleagues and friends
Interactive chain of thoughts
AI systems are often built to provide one-shot answers to user queries. However, researchers have shown that models are much more accurate when they structure their thinking and generate a chain of thoughts that defines the path to an answer for the user.
In addition, to provide only relevant results at the end of the search, products like Perplexity Copilot ask the user for clarifying information when the model in its chain of thoughts thinks it doesn't have all the information it needs to confidently continue its reasoning.
Perplexity Copilot eliminates the time-consuming process of asking the same questions of Google or ChatGPT over and over again and takes you directly to the answer if it exists. If it doesn't find an answer to the user's query, it will point that out instead of returning irrelevant results or hallucinating - it will confidently return incorrect results, making the user think the results are correct.
Multi-language support
Many tools based on generative AI can interact with users in the most popular languages. For example, you can ask ChatGPT a question in English, and ChatGPT will answer you in English. If you switch to French, ChatGPT will switch too.
Translation
Notion AI and other tools allow users to "translate" text into different languages.
Agent (or model / GPT), and model selection
Generative AI runs on agents or GPT applications from OpenAI stores that are based on AI models. Some AI products let the end user decide whether to use the GPT3.5 Turbo model or GPT4, which may be more expensive but more accurate, or between the YouTube and the Google Maps agent.
3. Personalization features
Change of tone
The "Change Tone" option in AI chatbots allows users to modify the style or emotional tone of the chatbot's responses, providing a customizable conversational experience. Common options include professional, casual, straightforward, confident, or friendly tone.
Change of style
In addition to changing tone, many chatbots allow users to choose from a variety of styling options, such as "shorter," "longer," and "simpler," to customize the length and complexity of the AI chatbot's responses.
Mimicking a role, character, or person
Let the user decide which role, character, or person they want the assistant to impersonate by setting the tone of your voice, the expressions the assistant uses, and more.
The most notable use case example is Character.AI, a leading conversational AI startup. Character.AI enables natural and contextual conversations with chatbots, and has attracted a massive user base of hundreds of millions worldwide. Character.AI's chatbots range from virtual assistants to interactive personas, allowing users to converse with legendary figures from history like Albert Einstein, the present like Elon Musk, or fictional characters like Goku or Barbie.
ChatGPT can also imitate characters like Shakespeare.
Writing assistants like Grammarly let you set your profession, such as product designer, to match the tone and words used more closely to what you need.
4. Explainability features
References to data sources, tools, and agents used
References, also mentioned in the section on Interactivity features, help users gain confidence in the results generated by artificial intelligence.
For example, Perplexity presents data sources through their colorful logos so that they stand out as the user searches for information. These are often logos from well-known data sources, such as Wikipedia or Britannica, or popular newspapers, such as the New York Times, so the user is left with little doubt about what Perplexity is providing.
Highlighted changes
DeepL highlights changed words in green, allowing for a quick scan of the corrected text to ensure it aligns with the user's expectations. This allows the user to easily and quickly review the changes made by the AI.
DeepL also provides the option to see what has been changed in what, further facilitating the review process.
Displaying the chain of thoughts
Having the AI model generate a chain of thoughts to get to an answer helps the model provide a more accurate answer and allows the user to understand and verify how the model is thinking.
Conclusion
I hope the article was insightful and useful. I will continue to write new articles that provide insight into the latest AI features and how to build them.
Subscribe to the newsletter on Substack to never miss updates.