OpenAI's Altman announces GPT-4 Turbo at first-ever developer conference


ChatGPT-maker OpenAI hosted its first developers conference today in downtown San Francisco announcing a new ChatGPT-4 Turbo model with increased knowledge through April 2023, and more.

A live-streamed keynote by CEO and co-founder Sam Altman kicked off the OpenAI DevDay event Monday at 1 p.m. PT, who also brought out Microsoft's CEO Satya Nadella in a surprise move to talk about how the tech companies partnership was going.

"Your mission and our mission is to empower every person on the planet...being able to get the benefits of AI to everyone,..and we're grounded in the fact that safety matters," Nadella said.

ADVERTISEMENT

"We commit ourselves deeply to making sure you all as builders of these foundation models have not only the best systems for training and inference but the most compute so that you can keep pushing forward on the frontier," he said.

"We have the best partnership in tech," Altman said. Altman also pointed out that a "gradual iterative deployment is the best way to address the safety challenges of AI... and move forward carefully."

The one-day event brought together 900 hundred developers from around the world “to preview new tools and exchange ideas,” OpenAI said, with more than 40,000 viewers on YouTube of the live conference.

100 million ChatGPT users weekly

Altman said there are now 100 million users of ChatGPT weekly since launching in November 2022. A huge jump from the record 100 million monthy users in January of this year.

"Eventually, you'll just ask the computer for what you need, and it'll do all of these tasks for you," Altman said.

That's besides the over 2 million developers who are using OpenAI’s latest large language model GPT-4, GPT-3.5, image generator DALL·E, and its automatic speech recognition (ASR) system Whisper, the company stated on its website.

ADVERTISEMENT

The conference is a way for the company to entice those developers, showing them how to best integrate OpenAI products into existing applications and entirely new ones – all while ensuring the use of best practices.

For example, besides fine-tuning GPT-3 to a specific task or domain by training it on custom data to improve its performance, developers can generate embeddings that can be used for tasks like text classification, search, and clustering, the company said.

A new Assistants API with new Retrieval, Code Interpreter, and Function Calling features will help developers build world-class AI assistants within their own apps that have goals and call models and tools, OpenAI said.

Altman said customization of ChatGPT’s will be improved, increased rate limits will be doubled, tokens will now be less than three times the original costs of GPT predecessors.

Additionally a Custom Models program, also launched Monday, will provide companies with a dedicated a group of reserachers to train and customize GPT-4 models directly for them.

OpenAI introduces GPTs
You can now create custom versions of ChatGPT that combine instructions, extra knowledge, and any combination of skills. Image by OpenAI.

Introducing GPT-4 Turbo

Altman quickly announced the launch of its newest model GPT-4 Turbo, supporting a 128K context window, reduced prices up to 50% for newer models and tools, as well as a slew of updates and improvements for release on Monday.

One of the biggest improvements that will positively affect both users and developers is that ChatGPT’s knowledge about the world will now be up to date through April 2023 as compared to September 2021.

The Turbo version of the large language model supports longer content length with more accuracy, higher control over model behavior, and better knowledge prompts new modalities, including a new natural and improved text-to-speech model.

ADVERTISEMENT

The 128K context window can fit the equivalent of more than 300 pages of text in a single prompt and the model boasts a new built-in Copyright Shield safeguard feature.

The Turbo model is available in preview to paying developers now, and a stable production-ready model will be released in the coming weeks.

Besides GPT-4 Turbo with Vision, upgraded models DALL·E 3 and Whisper will also now be available to developers.

Anyone can build a GPT

OpenAI also rolled out a new product line called GPTs, where individuals with no code experience can create custom versions of ChatGPT for a specific purpose and the ability to share it with others.

"Creating one is as easy as starting a conversation, giving it instructions and extra knowledge, and picking what it can do, like searching the web, making images, or analyzing data," the OpenAI team said.

OpenAI plans to open a GPT store in the coming months "featuring creations by verified builders" who can earn money based on usage and popularity.

The GPT store will have a searchable database of GPTs based on category. OpenAI said the store will also showcase a ranking leaderboard and "spotlight the most useful GPTs they come across.

Once formalities and the presentations are complete, OpenAI will encourage attendees to break off in smaller groups led by its technical staff to engage in more in depth discussions.

ADVERTISEMENT

The compay also said it would cover any legal costs regarding copyright infringement for its enterprse users, already offered to big business customers of Google and Microsoft.

OpenAI, founded in 2015, made news in October with new initiatives for its Frontier Model Forum along with Microsoft, Google, and the AI startup and development company Anthropic.

The four creators announced the AI safety forum’s first executive director Chris Meserole, along with a $10 million AI Safety Fund to promote responsible artificial intelligence research.

The OpenAI DevDay event will end with an evening reception.