OpenAI, the creator of ChatGPT, is developing a new method for its AI models under the project name “Strawberry,” per sources familiar with the matter and internal documents reviewed by Reuters.
Project Strawberry, which has not been previously disclosed, is part of OpenAI’s efforts to demonstrate that its models can achieve advanced reasoning capabilities as the company competes with other Microsoft-backed startups in the artificial intelligence space.
Reuters reported on “Strawberry,” previously known as Q*, last year as it was held up as a major breakthrough at the company. Two sources that were given demos of Q* earlier this year said the technology could solve complex science and mathematics problems that current commercially available models cannot handle.
OpenAI is focusing Strawberry on several capabilities, including the ability to perform long-horizon tasks (LHTs) which require models to execute a series of actions over an extended period of time.
OpenAI’s main goal is for their models to be able to use these skills to independently conduct research by browsing the internet. They plan to do this with the help of a human computer-using agent (CUA) that can make decisions based on its discoveries. Additionally, OpenAI intends to test their capabilities in performing tasks typically done by software and machine learning engineers.
