OpenAI's 'Strawberry': Advancing AI Reasoning Technology

13/07/2024 — OpenAI is developing a new AI project called “Strawberry,” aimed at significantly enhancing the reasoning capabilities of its models.

Overview

Internal documents and sources reveal that the project is a closely guarded secret, designed to enable AI to autonomously navigate the internet and conduct deep research. The project builds on previous efforts known as Q*, which demonstrated the ability to solve complex problems.

The Strawberry project aims to overcome current AI limitations in common sense and logical reasoning tasks. It involves a unique post-training process to refine the models after their initial training on large datasets. This technique is similar to Stanford’s Self-Taught Reasoner method, which allows AI to iteratively create its own training data and potentially surpass human intelligence.

Sam Altman Shaking Hands with Satya Nadella at OpenAI Event after he was ousted from his CEO role but subsequently joined Microsoft

OpenAI hopes Strawberry will enable its models to handle long-horizon tasks, requiring extended planning and execution. The project includes developing and testing models on a “deep-research” dataset, although specifics are not yet clear. OpenAI is signaling to developers that it is close to releasing this advanced technology, which could revolutionize AI’s ability to perform complex, autonomous tasks.

OpenAI's Strategic Focus and Industry Views

In addition to improving AI reasoning, OpenAI plans to use Strawberry for tasks like autonomous web research and software engineering.

Sam Altman, the CEO of OpenAI, a leading artificial intelligence research organization. Known for his visionary approach to AI

Other companies, including Google, Meta, and Microsoft, are also exploring ways to enhance AI reasoning, but opinions differ on the potential of large language models. OpenAI CEO Sam Altman and other researchers view advanced reasoning as crucial for future AI developments and applications.