OpenAI’s Secretive Project ‘Strawberry’ Aims to Enhance AI Reasoning

OpenAI is developing a new AI project under the code name “Strawberry,” according to sources cited by Reuters.

The agency reports that through Strawberry, the startup aims to significantly enhance the reasoning process of AI models, enabling them to plan actions in advance to “autonomously and reliably” gather information online and conduct “in-depth research.” The project involves a specialized method of tuning a neural network after its initial training on a large dataset.

Strawberry includes “post-training” of generative AI models. This refers to adapting and refining performance in a specific way after the training phase, the media notes.

The project’s capabilities include solving long-term complex tasks that require the system to plan ahead and execute a series of actions over an extended period.

The plan is described as “in development.” Work on Strawberry is kept under strict secrecy within the company, Reuters clarifies.

“We want our AI models to see and understand the world as we do. Continuous exploration of new AI capabilities is standard practice in the industry, and we are confident that over time these systems will improve their reasoning abilities,” commented an OpenAI representative.

Previously, the project was known as Q* and was considered a breakthrough for creating artificial general intelligence (AGI), which “could pose a threat to humanity.”

According to two Reuters sources, early in the year they saw demonstration versions of Q* capable of answering complex scientific and mathematical questions, which existing models cannot do.

In July, OpenAI developed a five-level scale to assess AI development, placing ChatGPT between the first and second levels.

Earlier, the company began collaborating with the laboratory that created the atomic bomb.