Raphael Thys
Nvidia and pharmaceutical startup Evozyne have debuted a generative AI model capable of producing proteins for use in medicine and other industries. The new Protein Transformer Variational AutoEncoder (ProT-VAE) is built on Nvidia’s BioNeMo framework and uses generative AI to rapidly create examples of synthetic protein designs that fit given parameters and speed up development. Nvidia and Evozyne have just published a paper with a couple of examples one that may cure a congenital disease and another that could consume carbon dioxide and combat global warming.
Protein Engineering AI
Nvidia started working with Evozyne, founded in 2020 by Paragon Biosciences, last year on designing ProT-VAE from BioNeMo. The idea is to apply deep learning and generative AI to biology the way it’s used for human language. The researchers concentrated on what’s known as the PAH gene, which instructs cells to produce phenylalanine hydroxylase, a digestive enzyme. If the gene isn’t coded correctly, people can develop phenylketonuria. The ProT-VAE model synthesized variants of PAH to see if it could produce a more effective version of the enzyme and potentially a genetic therapy. The model produced a new variant that was two and a half times better than the natural enzyme.
“We trained a model on protein data just like in other forms [of generative AI] but from amino acid sequences,” Nvidia vice president of healthcare Kimberly Powell told Voicebot in an interview. “With Evozyene, we evolved the model to create this new variant, which is the generative model. [The proteins] are put into families, and [researchers can] fine-tune the model for the specific family and not need a lot of data. At which point you can say, ‘hey, model, generate me a ton of new amino acid sequences never seen in nature before.'”
The researchers can use mathematical language to describe other parameters for the proteins, iterating and fine-tuning the model as much as they choose. It will then produce protein designs suitable for the described purpose, and researchers can comb through and pick out promising options to chemically synthesize for real-world testing.
“This is the evolution of methods to do protein engineering,” Powell said. “It’s putting biology into information science and turning it into engineering.
Generative Biology AI
Applying generative AI to biology is still a new idea but is on the cusp of rapid growth. Drug discovery startup Variational AI recently filed two provisional patents for two novel small molecules that could treat COVID-19 designed with generative AI on its Enki platform. Its variational autoencoder (VAE) was trained on a mix of public and proprietary datasets to produce the protease inhibitors, which could play a key role in future COVID-19 antiviral treatments. With Nvidia’s success in its partnership with Evozyne, this could become a standard approach to medical and biological research.
“This is an absolutely unprecedented way to understand biology. BioNeMo is in early access, but the traction is amazing. It’s all built on the shoulders of what we’ve accomplished with natural language processing,” Powell said. “The network architecture is different, and how you use the model in biology is very different from ChatGPT or language use cases, but it’s all about understanding relationships. We’re still just scratching the surface.”
Generative AI Produces COVID-19 Treatments Patented by Drug Discovery Tech Startup Variational AI
OpenAI may soon charge for a version of its popular generative AI chatbot ChatGPT. The startup announced its plans on Discord with a link to sign up for the waiting list of what it calls ChatGPT Professional with a series of questions related to the cost and use of an enterprise version of the chatbot.
ChatGPT Professional
OpenAI explained in its announcement that the company is “starting to think about how to monetize ChatGPT” so that it can maintain its “long-term viability.” ChatGPT Professional is supposed to address those issues, with the survey explaining that the still experimental product will always be available, have a fast response, and have at least twice the current limit of messages so that the user doesn’t run out the way it’s possible to do with the current free version. The survey asks questions about how the applicant would use ChatGPT professionally and what monthly price they would consider too high.
“If you are selected, we’ll reach out to you to set up a payment process and a pilot,” OpenAI explains in the survey. “Please keep in mind that this is an early experimental program that is subject to change, and we are not making paid pro access generally available at this time.”
ChatGPT is likely to find plenty of people willing to pay any reasonable price, especially if the free version of the massively successful tool is cut back or eliminated in favor of the upgraded service. Demand has already prompted Microsoft to work on integrating ChatGPT into Microsoft Word and other Office products as well as its Bing search engine. It will also supposedly deepen its $1 billion investment in OpenAI to $10 billion.
Microsoft plans to incorporate OpenAI’s ChatGPT and upcoming GPT-4 into Microsoft Word and other Office products, according to a report from The Information. The plans to do so follow similar news about integrating ChatGPT into Microsoft’s Bing search engine and will go along with a potentially massive jump in the tech giant’s already enormous $1 billion investment in OpenAI.
Office ChatGPT
The report explains that Microsoft wants to bring the power of generative AI to Word, PowerPoint, Outlook, and the entire suite of Office products so that users can generate text appropriate for their projects based on a short prompt. Microsoft also wants to personalize the AI, training it on what a particular user writes though without compromising their privacy and sharing their data. The use and interst in generative AI for images and text has already led to the creation of many AI writing assistants for business and persona use. Augmenting word-processing and other business software applications with the technology was a logical extension of that effort.
“How Microsoft intends to use these capabilities is not entirely clear. Companies like Jasper AI have had great success offering full AI-enabled writing suites with numerous templates that generate entire documents for business professionals. On the other end of the spectrum, there are tools like Grammarly and AI21’s Wordtune that don’t write for the user but do make suggestions on how to improve the writing,” Voicebot founder Bret Kinsella commented in the Synthedia newsletter.
Microsoft has already embedded OpenAI’s DALL-E 2 text-to-image generator into its Azure OpenAI Service, as well as the new Designer app and the new Image Creator tool for Bing and Edge. The generative text word is filling up with tools on major platforms like Canva, as well as independent products like AI21’s Wordtune. The chatbot format of ChatGPT is also drawing imitators for help with writing and other tasks, such as Jasper Chat, a chatbot format for Jasper’s own generative AI engine.
Generative content startup Jasper has released a chatbot version of its generative AI engine with an eye toward business use. The new Jasper Chat responds to direct requests to complete tasks and is likely to leverage the burst of interest in OpenAI’s ChatGPT chatbot but tuned specifically for business, marketing, and finance use cases.
Jasper Chat
Like ChatGPT and other generative AI chatbots, Jasper Chat simplifies the process of producing information by turning the programming into a conversation with natural language, mimicking human interaction with an assistant. Users can ask Jasper Chat to compose text, suggest variations or edit existing text, and further refine the results until the user is satisfied.
“We’re living through a time where generative AI is widely available, but not easily accessible to complete mission critical tasks,” Jasper CEO Dave Rogenmoser said. “With Jasper Chat, we’ve harnessed the power of language learning models and made the interface feel as natural as possible for business users. Jasper Chat is more of a new way to interact with existing AI than a new technology itself, but it has the potential to make generative AI even more widely adopted than it already is.”
Jasper debuted last year with its own generative AI engine, not relying solely on OpenAI’s GPT-3 and related models, but instead using a mixture of model sources. The company looks to produce content based on limited suggestions for businesses that need social media posts, website content, long-form articles, and other kinds of written content. On the consumer side, Jasper offers help writing and editing text and images. The Jasper Google Chrome extension enables users to generate content that fits in the context of their open tabs. Jasper’s main target is in business, however, and the company fine-tunes its AI for that purpose, with the ability to adapt it to different customers. Jasper Chat continues that focus, with a streamlining feature of up and down-votes for any result, allowing Jasper to better learn what it users are looking for. The company claims to have more than 100,000 business users currently, and raised $125 million in October.
NB: Updated following Voicebot’s email interview with Dogtown Media CEO Marc Fischer.
Mobile app developer Dogtown Media has introduced what it claims is the first iOS and Android app “powered by ChatGPT,” OpenAI’s enormously popular generative AI chatbot. The free Merlin AI app says it offers a way to interact with the ChatGPT AI, without navigating to the website. However, OpenAI hasn’t yet offered an API for ChatGPT. Unless OpenAI gave Dogtown Media special access to an otherwise unavailable API, the app acts as a messenger between the user and ChatGPT’s online home, or the app uses GPT-3 or GPT-3.5 API and the marketing is just capitalizing on ChatGPT’s popularity. We’ve reached out to Dogtown for more information.
Update: Dogtown Media CEO Marc Fischer clarified in an email interview with Voicebot that the company used OpenAI’s GPT-3.5 API. “The app is able to instantaneously run complex queries and return the highest caliber results that the AI tech stack can provide.” GPT-3.5 is the same baseline model as ChatGPT, hence the marketing copy calling it the first mobile app for ChatGPT.
ChatGPT Mobile
At first glance, the Merlin AI app is a very straight recreation of ChatGPT with a simple blank area to type requests. One notable difference is that the response comes all at once in a block rather than simulating a very fast typer the way ChatGPT performs on the website. The conversational element is there, too as the AI responds in the context of the ongoing discussion, and the entire conversation is visible in the app. Dogtown envisions Merlin AI’s mobile ChatGPT as a potential replacement for Google search, using conversation instead of links to other pages with answers.
“To be honest, ChatGPT is the best-of-breed AI model and miles ahead of everything else on the market,” Fischer told Voicebot. That being said, we’re optimistic about what the future holds. In a short amount of time we’ve built a budding relationship with OpenAI’s team, which we’ll expand into exciting new territories into the new year.”
Update: Voicebot founder Bret Kinsella deduced how Merlin works before the company’s confirmation. His analysis of the benefits and potential problems of trying to make a ChatGPT mobile app without an API mirror the reasons behind Merlin’s structure.
“I am not aware of an actual ChatGPT API from OpenAI. More likely, they are using the GPT-3 API that is represented as a chat interface. They may even be using the GPT 3.5, which is the model ChatGPT employs. Jasper AI showed something in similar to this earlier in the week,” Kinsella commented earlier. “The API integration would be the most logical approach but suggests the marketing claims are a little misleading and using the ChatGPT name because it is recognizable.”
If Merlin AI used ChatGPT, it would need to integrate through the website. That would mean sending the user’s input and then scraping the response and reposting it on the app. While that would explain the marketing, it comes with its own potential problems.
“OpenAI positions ChatGPT as a beta that could be changed or removed at any time,” Kinsella said. “You don’t want to have your mobile app suddenly become non-functional because OpenAI changed the access status for ChatGPT. Regardless, I am excited to see this type of app hit the market. It will expose more people to the functional benefits of large language models and has a super user-friendly interface which will reduce adoption friction.”
Merlin AI’s speedy launch just weeks after ChatGPT appeared is remarkable as Dogtown would have needed to develop and then navigate the app through the Android and iOS publishing systems at a brisk clip. Dogtown has previously published more than 250 mobile apps for healthcare, finance, and other industries. That might spark some skepticism on its provenance.
Update: The speediness still applies in terms of the GPT-3.5 model, as Fischer boasted that Merlin debuted just eight days after the initial proposal. There are also plans for expanding and enhancing the app in the future.
“My team at Dogtown Media (the backer of Merlin AI) have been developing mobile apps since the introduction of the Apple App Store and I’m proud to say that this is by far our fastest app launch on record,” he wrote. “The app you see in the App Stores now is only the tip of the iceberg. The Merlin AI team is working around the clock to add advanced functionality and will be scaling up to include ChatGPT-4 capabilities when they’re released next year. Keep an eye out for a flurry of new premium features being added over the coming weeks and months.”