OpenAI is developing an artificial intelligence system using a novel approach, codenamed Strawberry. This new model is distinguished by its advanced reasoning capabilities. According to Reuters, this information comes from an internal OpenAI document reviewed by the agency’s journalists in May.
Project Details and Capabilities
The exact date of the document remains undetermined, but it outlines how OpenAI plans to use Strawberry for research purposes. The model is currently under development, as a source informed the publication. The timeline for Strawberry’s release to the public is unclear, and the project remains highly classified, with access strictly guarded within OpenAI. The document describes Strawberry’s potential not only to provide answers to questions but also to create autonomous AI navigation plans for performing “deep research” on the Internet.
Internal Confirmation and Development History
OpenAI has acknowledged the project without denying its existence. A company spokesperson stated, “We want our AI models to see and understand the world the same way we do. Continuous exploration of new AI capabilities is common practice in the industry, and we share confidence that these systems will improve their reasoning abilities in the future.” Work on the project began last year under the name Q*, and it faced challenges, including the dismissal of Sam Altman shortly after its launch. Two OpenAI employees witnessed demonstrations of Q*’s capabilities this year, where the model answered complex scientific questions and solved mathematical problems.
On Tuesday, the company held an internal general meeting, showcasing an AI with human-like reasoning skills. While an OpenAI representative confirmed the meeting, they did not disclose specific details, and Reuters could not confirm if it pertained to the Strawberry project. We’ll keep you updated as more information becomes available.
Anticipated Impact and Technological Advancements
In recent months, OpenAI has privately indicated to developers and third parties that it is close to releasing technology related to significantly advanced AI reasoning abilities, according to anonymous sources. A unique feature of Strawberry is its post-training processing method, often referred to as “fine-tuning.” This method shares similarities with the StaR (Self-Taught Reasoner) approach, developed in 2022 at Stanford University. StaR involves self-learning AI that iteratively prepares its own data sets for additional training, potentially creating an AI model that surpasses human-level intelligence.
Strawberry’s most notable capability is performing tasks that require long-term planning and executing a series of actions over an extended period, notes NIX Solutions. OpenAI is training and evaluating models using “deep research” data, though the specific composition and duration of this data set remain unclear. The model autonomously conducts its research projects by surfing the web with a special agent known as a Computer-Using Agent (CUA). As part of the verification process, the model will perform tasks assigned to software and machine learning engineers.
OpenAI’s Strawberry project represents a significant step forward in AI technology, promising enhanced reasoning abilities and autonomous research capabilities. We’ll keep you updated on any further developments.