OpenAI is working on a new approach to its artificial intelligence models as part of a project codenamed Strawberry. Reuters reports this with reference to a person familiar with the matter and internal documentation.
According to him, the way Strawberry works is a strict secret even within OpenAI. But the documentation still reveals some of its details.
The project is ostensibly focused on ensuring that AI models not only generate answers to queries, but are also capable of planning ahead for autonomous navigation on the Internet - to perform what OpenAI calls "deep exploration."
According to more than a dozen artificial intelligence researchers, this is something that AI models have failed to do so far.
At the same time, the publication draws attention to the fact that the Strawberry project was previously known as Q* (pronounced Q-Star), which became known last year. Back then, the project was called a breakthrough in the creation of general artificial intelligence (AGI).
Two sources said that earlier this year they saw what OpenAI staff called a demo version of Q*, which is capable of answering complex scientific and mathematical questions beyond the reach of today's models.
During a recent internal meeting, OpenAI demonstrated a research project that claims to have new human-like reasoning skills.
An OpenAI representative confirmed the meeting but did not provide details of its content. The publication was unable to determine whether Strawberry's project was demonstrated.