By William J. Furney
ChatGPT may no longer “hallucinate” as much and give more accurate answers, especially to demanding questions, as the maker of the wildly popular virtual assistant has introduced new models that “think like a person” before they start talking.
Since its launch almost two years ago, in November 2022, and its lightning-fast use by tens of millions of people — from students wanting papers written to doctors looking for diagnoses — the artificial intelligence (AI) app has become notorious for making things up.
Maker OpenAI, backed by billions in Microsoft funding, acknowledged early on that its product was not always accurate, and continues to warn that “ChatGPT can make mistakes. Check important info.”
But this week OpenAI unveiled two new models that won’t rush into answering users’ queries, joining a stable of five that are both free and paid-for, at $20 per month. The new arrivals are called o1, for “advanced reasoning” and currently in preview mode, and o1-mini, which is “faster at reasoning”.
“We trained these models to spend more time thinking through problems before they respond, much like a person would,” OpenAI said in a statement. “Through training, they learn to refine their thinking process, try different strategies and recognise their mistakes.”
OpenAI said that during testing, “the next model update performs similarly to PhD students on challenging benchmark tasks in physics, chemistry and biology. We also found that it excels in math and coding.”
The San Francisco-based company headed by AI pioneer Sam Altman said that previous ChatGPT models, such as 4o, mostly failed to correctly answer questions in an exam for the International Mathematical Olympiad, an annual academic competition, scoring only 13 percent. But the new versions got 83 percent of the exam questions right.
Instead of the instant answers previous versions give, the new ones take anywhere from a few seconds to half a minute before giving a result, displaying a “thinking” notice while sorting through relevant information in its vast language database and deciding how best to answer.
For “complex reasoning tasks this is a significant advancement and represents a new level of AI capability,” said OpenAI.
It said the “enhanced reasoning capabilities may be particularly useful if you’re tackling complex problems in science, coding, math and similar fields.
“For example, o1 can be used by healthcare researchers to annotate cell sequencing data, by physicists to generate complicated mathematical formulas needed for quantum optics and by developers in all fields to build and execute multi-step workflows.”
The new models do not yet have the internet-searching and image-generation capabilities of previous ones, however, OpenAI said.
- Image credit: Furney Times