The long-awaited follow-up to ChatGPT has gone live, boasting of “human-level performance” in university-standard exams.
OpenAI said GPT-4, the next generation of its artificial intelligence-powered chatbot, marked a “milestone” in the development of deep learning, which imitates how humans gain knowledge.
“We’ve spent 6 months iteratively aligning GPT-4 using lessons from our adversarial testing program as well as ChatGPT, resulting in our best-ever results (though far from perfect) on factuality, steerability, and refusing to go outside of guardrails,” the San Francisco-based company said in a blog post on Tuesday.
OpenAI, which is backed by Microsoft, said the new version of its AI-powered chatbot is a “multimodal” model that can generate content from both images and text prompts.
In an online demonstration, OpenAI President Greg Brockman showed GPT-4 creating a real website based on a hand-drawn mock-up.
OpenAI said the update is able to pass the bar exam for prospective lawyers with a score in the top 10 percent of applicants, compared with the bottom 10 percent of test-takers previously.
The chatbot can also beat 90 percent of humans who take the evidence-based reading and writing section of the Scholastic Assessment Test and the verbal section of the Graduate Record Examination used for admission to postgraduate education, OpenAI said.
GPT-4 is also much less likely to produce inaccurate, offensive, insulting answers than ChatGPT, the company said.
“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.
OpenAI, however, said GPT-4 is still “not fully reliable” and can still produce unexpected answers known as “hallucinations” and reasoning errors.
OpenAI’s launch of ChatGPT in November took the tech world by storm, prompting existential questions about the future of sectors ranging from education to journalism and healthcare.
Tech giants including Google, Microsoft, Huawei, Alibaba, and Baidu are racing to roll out their own versions of the technology amid heated competition to dominate the burgeoning AI sector.
