OpenAI is hosting a developer live stream that will show the first public demo of ChatGPT-4. The new Large Language Model (LLM) has reportedly been in development for a few years, and Microsoft confirmed it’s the tech powering the company’s new Bing Chat service.
The presentation starts at 1 p.m. PT on Monday, March 14. OpenAI President and co-founder Greg Brockman will lead the presentation, walking through what GPT-4 is capable of, as well as its limitations. You can see a live stream of the event below.
OpenAI has already announced that ChatGPT-4 will only be available to ChatGPT Plus subscribers. The free version of ChatGPT will continue to run on the GPT-3.5 model.
GPT-4: new features, visual input, availability, and more
Edge Copilot finally delivers on Microsoft’s Bing Chat promises
Hackers are using AI to spread dangerous malware on YouTube
The live stream is focused on how developers can leverage GPT-4 in their own AI applications. OpenAI has recently made its API available to developers, and companies like Khan Academy and Duolingo have already announced that they plan on using GPT-4 in their own apps.
Although speculation has been wild for what GPT-4 could be capable of, OpenAI is describing it as an evolution of the existing model. The new model will be able to mimic a particular writing style more closely, for example, as well as process up to 25,000 words of text from the user.
OpenAI says that ChatGPT-4 doesn’t need text, either. It can receive an image as a prompt and generate a response based on it.
Otherwise, the new version has updated security features. OpenAI claims it’s 82% less likely to offer disallowed responses, and it provides 40% more factual responses. It’s tough to say what that means in practice at the moment, however.
Although the new model could vastly expand the capabilities of ChatGPT, it also comes with some worries. Microsoft’s Bing Chat has already shown some unhinged responses, and it uses the GPT-4 model. OpenAI warns that the new model could still have these issues, occasionally showing “social biases, hallucinations, and adversarial prompts.”