OpenAI stated on Thursday it was launching its “Strawberry” sequence of AI models designed to spend extra time processing solutions to queries so as to clear up arduous issues.
The models are able to reasoning by way of advanced duties and might clear up more difficult issues than earlier models in science, coding and math, the AI agency stated in a weblog put up.
OpenAI used the code title Strawberry to refer to the mission internally, whereas it dubbed the models introduced on Thursday o1 and o1-mini. The o1 might be accessible in ChatGPT and its API beginning Thursday, the corporate stated. ChatGPT has struggled to recognize that the phrase “strawberry” comprises three situations of the letter R.
Noam Brown, a researcher at OpenAI centered on bettering reasoning within the firm’s models, confirmed in a put up on X that the models have been the identical because the Strawberry mission.
“I’m excited to share with you all of the fruit of our effort at OpenAI to create AI models able to actually common reasoning,” Brown wrote.
In its weblog put up, OpenAI stated the o1 mannequin scored 83% on the qualifying examination for the Worldwide Arithmetic Olympiad, in contrast with 13% for its earlier mannequin, GPT-4o.
The mannequin additionally improved efficiency on aggressive programming questions and exceeded human PhD-level accuracy on a benchmark of science issues, the corporate stated.
Brown stated the models have been in a position to accomplish the scores by incorporating a approach generally known as “chain-of-thought” reasoning, which entails breaking down advanced issues into smaller logical steps.
Researchers have famous that AI mannequin efficiency on advanced issues tends to enhance when the method has been used as a prompting approach. OpenAI has now automated this functionality so the models can break down issues on their very own, with out consumer prompting, the corporate claimed in its weblog put up.
“We educated these models to spend extra time considering by way of issues earlier than they reply, a lot like a particular person would. Via coaching, they study to refine their considering course of, strive completely different methods, and acknowledge their errors,” OpenAI stated.