OpenAI revealed a lot of significant innovations during its first-ever developer conference on Monday. The company reported it currently has two million developers and 100 million weekly active users, with 92% of Fortune 500 companies building on its products. OpenAI is the “most advanced and widely used AI platform in the world,” said CEO Sam Altman during his keynote presentation. Here are the most important announcements from DevDay.
ChatGPT-4 Turbo
OpenAI kicked off its DevDay with a bang introducing a new GPT-4 Turbo model, delivering improved function calling, knowledge, lowered pricing, new modalities, and more. ChatGPT-4 was already the most advanced model on the market, but Turbo is significantly more powerful. Turbo supports 128,000 tokens of context, 16x more context than its predecessor could handle. That means you could enter 300 pages of a book and a request such as “summarize this book” into ChatGPT, and it could handle it. That example is just the tip of the iceberg, and the increased context window is a significant improvement for large language models.
GPTs
You’ve heard of ChatGPT, but OpenAI is launching GPTs: a tailored version of ChatGPT for specific purposes. You can build a GPT for almost anything, simply by talking to a chatbot. For example, Canva built a GPT where users can talk to a chatbot powered by ChatGPT to design posters. Users can ask the Canva GPT, “Can you build me a poster for an 80’s Halloween party?” and the GPT will build you a poster on Canva with your request. You can customize options like the date, location, style, and tone simply by asking the chatbot. These are a new concept for OpenAI, but they’re investing big into hyper-customized chatbots like the Canva example.
GPT Store
The GPT announcement comes alongside a new GPT store that looks much like the App Store or Google Play. Users can browse different GPTs for various use cases. Perhaps you could download a chatbot that would help you build posters like the Canva example. Another example could be a chatbot uploaded with the lectures of a busy professor, enabling students to ask questions about their class without going to office hours. The possibilities are truly endless for this GPT store. Not to mention, OpenAI announced they will share revenue with developers who create the most successful GPTs.
New Knowledge Cutoff
Sam Altman announced that ChatGPT products will now have a new knowledge cutoff of April 2023. The old knowledge cutoff was September 2021, a significant limitation of the world’s leading generative AI company. Altman said the company “will try not to let it get that out of date again.” The knowledge cutoff is a limitation that Elon Musk’s Grok, launched last weekend, does not have. Because Grok is connected to X’s live tweets, there is no knowledge cutoff. However, this is probably the one good thing about Grok compared to GPT-4 Turbo, as Grok is significantly limited in nearly every other regard.
Custom Models Program
OpenAI is launching “Custom Models,” a program giving selected organizations an opportunity to work with a group of dedicated OpenAI researchers to train custom ChatGPT-4 models specifically for their work. OpenAI researchers can now work with specific companies, non-profits, or government agencies, to build a perfect GPT-4 for their work, and these organizations will have exclusive access to their models. The proprietary data will not be reused in any other context. It’s an opportunity to help organizations that don’t understand large language models create something meaningful with OpenAI’s tools.
ChatGPT Gets a Facelift
The ChatGPT that most people interact with (the consumer-facing model) received a new, cleaner user interface. OpenAI heard numerous complaints about its clumsy dropdown menu on the home screen, so it removed it completely. The website now looks less like a computer science project and more like a consumer product. ChatGPT for consumers will also now run on GPT-4 Turbo, the latest and most powerful version of ChatGPT.
OpenAI Lowers Prices
The company announced lower prices across the board on Monday, making it easier for developers to build on the platform. Altman shared that GPT-4 Turbo will be 2.75x cheaper than ChatGPT-4, and all older models will be significantly reduced in price as well.
Retrieval Connects Your Cloud to ChatGPT
Users can now upload their own data to work within ChatGPT’s model. Companies could upload their Google Drive or Microsoft 365 accounts so that all emails, documents, product information, presentations, and more are shared with ChatGPT, giving the chatbot work assistant capabilities similar to Microsoft 365 Copilot.
Satya Nadella to OpenAI: “We Love You Guys”
Microsoft CEO Satya Nadella made an appearance at the conference, saying the partnership with OpenAI has been fantastic and expressing his love for Altman’s company.
“The shape of Azure has rapidly changed and is changing rapidly in support of these models you’re building,” said Nadella.
Microsoft’s CEO says Azure is working to build the best systems to support these products he believes will benefit people worldwide. “Ultimately, it’s about being able to get the benefits of AI broadly disseminated to everyone,” he said.
Assistants API, Copyright Shield, and Other Developer Specific Improvements
OpenAI launched Assistants API on Monday to help developers build GPTs within their own applications. The launch provides several tools that will take a lot of the heavy lifting off of developers. Developers can also integrate AI image generation from DALL-E 3, along with text-to-speech functionality, directly into applications they build. Furthermore, Copyright Shield will protect developers who run into copyright infringement cases using OpenAI products in a court of law, a protection that is now commonplace in the AI world. The new tools from OpenAI are now cheaper than ever, and easier to use.