WITH ‘HUMAN-LEVEL’ PERFORMANCE, OPENAI UNVEILS CHATGPT SUCCESSOR

Wed 15 March 2023:

The long-awaited sequel to ChatGPT is now available, boasting of “human-level performance” in university-level exams.

GPT-4, the next generation of OpenAI’s artificial intelligence-powered chatbot, was described as a “milestone” in the development of deep learning, which mimics how humans learn.

“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.

PODCAST: THE HUMAN COST OF CHATGPT

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.

NYC EDUCATION DEPARTMENT BLOCKS USE OF THE CHATGPT AMID CHEATING WORRIES

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.