The latest version of ChatGPT is now available via a waitlist.
“It is more creative than previous models, it hallucinates significantly less and it is less biased,” Altman said in the thread. “It can pass a bar exam and score 5 on several AP exams.”
According to a page devoted to GPT-4 on the company’s website, ChatGPT-4 scored in the top 10% of test-takers on a simulated bar exam, while its predecessor, ChatGPT-3.5, scored in the bottom 10%.
At the same time, Altman cautioned that ChatGPT-4 is “still flawed, still limited, and it still seems more impressive on first use than it does after you spend some more time with it.”
With this newest model, OpenAI is previewing visual input for ChatGPT-4 and allowing developers to significantly customize its behavior, Altman said in the thread.
According to the webpage, the new model can accept image and text inputs while emitting text outputs. The image inputs are not yet publicly available, though; they are only in a research preview.
The firm is also open-sourcing its framework for automated evaluation of artificial intelligence (AI) performance, to allow anyone to help OpenAI improve its models, Altman added.
On its website, OpenAI said it has been using ChatGPT-4 internally in functions like support, sales, content moderation and programming.
“We have had the initial training of GPT-4 done for quite a while, but it’s taken us a long time and a lot of work to feel ready to release it,” Altman said in the thread. “We hope you enjoy it and we really appreciate feedback on its shortcomings.”
The previous version of ChatGPT captured the public imagination and took social media by storm in December as consumers flocked to play with the free AI chatbot’s humanlike conversation capabilities.
“We see this as an industrial revolution for knowledge work, and a moment where all our clients will need to rethink their business architectures and adapt,” Bain & Company Worldwide Managing Partner Manny Maceda said at the time.