ChatGPT maker claims its new AI model can reason and think "as a person"

() – OpenAI has unveiled a new artificial intelligence model that it claims can “reason” and solve more difficult scientific, coding and math problems than its predecessors.

The model, the first in a series called OpenAI o1, was unveiled Thursday as a pre-release, and the company says it expects regular updates and improvements. It will gradually become available to most ChatGPT users.

“We have trained these models to spend more time thinking about problems before responding, as a person would do,” explains the manufacturer of ChatGPT on its website. “Through training, they learn to refine their thought process, try different strategies and recognize their mistakes.”

As examples of the power of the new models, OpenAI noted that they can be used by health researchers to annotate cell sequencing data and by physicists to generate “complicated mathematical formulas needed for quantum optics.”

Noam Brown, a research scientist at the company, also highlighted the potential of the new AI models. “OpenAI’s o1 thinks for seconds, but our goal is for future versions to think for hours, days, and even weeks. Inference costs will be higher,” public at X on Thursday, referring to the costs — such as higher energy bills — of using AI to make inferences from input data. “But what cost would you pay for a new cancer drug? For innovative batteries?” he asked.

On Thursday, senior White House officials and top US technology executives were set to discuss AI’s massive thirst for power. Sam Altman, CEO of OpenAI; Ruth Porat, a top Google executive; and Dario Amodei, CEO of Anthropic, were expected to attend the meeting, a person familiar with the matter told .

While technology can help solve thorny problems like cancer and the climate crisis, it poses equally complex challenges, including how to meet the massive electricity demands required by advanced AI systems, which could worsen global warming.

According to the company, the new OpenAI model does not yet have many of the functions “that make ChatGPT useful,” such as browsing the Internet for information and uploading files and images. “But for complex reasoning tasks, it represents a significant advance,” it adds.

In tests, OpenAI o1 performs similarly to PhD students on difficult physics, chemistry and biology tasks, the company says. And in a qualifying exam for the International Mathematical Olympiad, the new series of models solved 83% of problems correctly.