OpenAI has created GPT-4, the latest milestone in OpenAI’s effort in scaling up deep learning.
“The AI technology “exhibits human-level performance on some professional and academic tasks,” a company blog noted.
The company said the model is “more creative and collaborative than ever before” and would “solve difficult problems with greater accuracy” than its earlier versions.
With its update, text responses from GPT-4 will be more accurate, and — in future — will come from both image and text inputs in a major leap forward for the technology, though this aspect has not yet been released.
For example, if a user sends a picture of the inside of a refrigerator, GPT-4 will not only correctly identify what is there, but also concoct what can be prepared with those ingredients.
OpenAI said it was working with a partner company, Be My Eyes, to prepare the next advance.
Much of the new model’s firepower is now available to the general public via ChatGPT Plus, OpenAI’s paid subscription plan and on a AI-powered version of Microsoft’s Bing search engine that is currently being tested.
OpenAI is backed by Microsoft, which earlier this year said it would finance the research company to the tune of billions of dollars.
The Windows-maker then swiftly integrated the tech into its Bing search engine, Edge browser and other products.
Microsoft’s aggressive adoption of ChatGPT has sparked a race with Google which announced its own versions of the AI technology, with Amazon, Baidu and Meta also wading in, eager to avoid being left behind.
OpenAI said that the new version was far less likely to go off the rails than its earlier chatbot with widely reported interactions with ChatGPT or Bing’s chatbot in which users were presented with lies, insults, or other so-called “hallucinations.”
“We spent six months making GPT-4 safer and more aligned. GPT-4 is 82 percent less likely to respond to requests for disallowed content and 40 percent more likely to produce factual responses,” OpenAI said.
Founder Sam Altman admitted that despite the anticipation, GPT-4 “is still flawed, still limited, and it still seems more impressive on first use than it does after you spend more time with it.”