ChatGPT Maker Unveils New AI Model That Can Reason Like a Human

OpenAI has introduced a new artificial intelligence model, described as more capable of “reasoning” and solving complex problems in areas such as science, coding, and math compared to its previous versions. This new model, named OpenAI o1, was released on Thursday as a preview, with the company planning regular updates. It will soon be available to most ChatGPT users.

In a statement, OpenAI explained, “We’ve trained these models to spend more time thinking through problems before providing a response, similar to how a person would. Through this training, they refine their problem-solving methods, try different strategies, and identify their mistakes.”

The company provided examples of how the new model could be applied, noting that it can assist healthcare researchers in annotating cell sequencing data or help physicists generate complex mathematical formulas for quantum optics.

Noam Brown, a research scientist at OpenAI, emphasized the potential of the o1 model. “OpenAI’s o1 can think for seconds, but future versions could think for hours, days, or even weeks,” he shared on social media. Brown acknowledged the higher energy demands associated with these advancements but posed the question, “What cost would you pay for a new cancer drug or breakthrough batteries?”

On Thursday, senior White House officials were set to discuss the energy demands of AI with tech leaders, including OpenAI CEO Sam Altman, Google’s Ruth Porat, and Anthropic CEO Dario Amodei, according to a source familiar with the matter.

While AI holds the promise of addressing significant challenges like cancer and climate change, its massive energy consumption presents its own issues, potentially exacerbating global warming.

OpenAI noted that its new model currently lacks some of the functionalities that make ChatGPT practical, such as web browsing and file or image uploads. However, it highlighted that the model represents a substantial improvement for tackling complex reasoning tasks.

In trials, the OpenAI o1 model performed at a level comparable to PhD students on challenging benchmarks in fields like physics, chemistry, and biology. Additionally, in a qualifying exam for the International Mathematics Olympiad, the model successfully solved 83% of the problems presented.