OpenAI Secretly Working on Project ‘Strawberry’ to Enhance Reasoning and Build Autonomous AI Agents

OpenAI, the creator of ChatGPT, is reportedly working on a new AI technology under the code name "Strawberry." This project aims to significantly enhance the reasoning capabilities of its AI models, as revealed by internal documents and a source familiar with the development.

The project's specifics, which have not been previously disclosed, involve a novel approach that allows AI models to plan ahead and navigate the internet autonomously to perform in-depth research. This advancement could address current limitations in AI reasoning, such as common sense problems and logical fallacies, which often lead to inaccurate outputs.

Project Strawberry was previously known as Project Q*, which was leaked last year and was capable of solving previously unseen math problems.

OpenAI's teams are working on Strawberry to improve the models' ability to perform long-horizon tasks (LHT), which require planning and executing a series of actions over an extended period. 

The project involves a specialised "post-training" phase, adapting the base models for enhanced performance. This method resembles Stanford's 2022 "Self-Taught Reasoner" (STaR), which enables AI to iteratively create its own training data to reach higher intelligence levels.

A spokesperson from OpenAI acknowledged ongoing research into new AI capabilities but did not directly address the specifics of Strawberry. The internal document indicates that Strawberry includes a "deep-research" dataset to train and evaluate the models, although the contents of this dataset remain undisclosed.

In recent months, OpenAI has privately hinted at releasing technology with advanced reasoning capabilities, aiming to overcome challenges in AI research and development. 

This innovation is expected to enable AI to conduct research autonomously, using a "computer-using agent" (CUA) to take actions based on its findings. Additionally, OpenAI plans to test Strawberry's capabilities in performing tasks typically done by software and machine learning engineers.

OpenAI has recently unveiled a five-level classification system to track progress towards achieving artificial general intelligence (AGI) and superintelligent AI. OpenAI executives shared this classification system with employees during an internal meeting and plan to share it with investors and external parties. The company currently considers itself at Level 1 and anticipates reaching Level 2 in the near future.

Other tech giants like Google, Meta, and Microsoft are also exploring techniques to enhance AI reasoning. However, experts like Meta's Yann LeCun argue that large language models may not yet be capable of human-like reasoning.

OpenAI's CEO, Sam Altman, emphasised earlier this year that reasoning ability is crucial for AI progress. The Strawberry project could mark a significant step towards AI models achieving human or super-human-level intelligence, potentially revolutionizing how AI assists in scientific discoveries and software development.