SAN FRANCISCO, CA – ChatGPT creator OpenAI is working on an innovative project codenamed “Strawberry,” designed to enhance the reasoning capabilities of its artificial intelligence models, according to sources familiar with the matter and internal documentation reviewed by Reuters.
The confidential project marks a significant step for the Microsoft-backed startup as it strives to demonstrate that its AI models can perform advanced reasoning tasks. Details of Strawberry, not previously reported, emerged from an internal OpenAI document seen in May. The exact date of the document remains unclear, but it outlines OpenAI’s plans to leverage Strawberry for conducting deep research. The project is still in progress, and the timeline for its public release is uncertain.
Inspired by Stanford’s ‘Self-Taught Reasoner’
The workings of Strawberry are tightly guarded secrets within OpenAI. The project bears resemblance to a 2022 method from Stanford called ‘Self-Taught Reasoner’. Strawberry aims to enable AI to autonomously and reliably navigate the internet and perform “deep research,” moving beyond generating simple answers to planning ahead.
More than a dozen AI researchers interviewed by Reuters confirmed that such autonomous reasoning remains a challenge for current AI models. An OpenAI spokesman acknowledged the continuous research but did not directly address questions about Strawberry.
Previously Known as Q*
Strawberry was formerly referred to as Q*, which Reuters reported last year as a breakthrough within the company. Earlier this year, sources witnessed demos of Q* that showcased its ability to tackle complex science and math problems beyond the reach of commercially available models.
In a recent internal meeting, OpenAI demonstrated a research project with purported human-like reasoning skills, as reported by Bloomberg. An OpenAI spokesman confirmed the meeting but declined to provide specifics on the demonstration, leaving it unclear if Strawberry was involved.
Advancing AI Reasoning Capabilities
OpenAI aspires for Strawberry to significantly enhance the reasoning abilities of its AI models. This involves a specialized post-training process after the AI model has been pre-trained on extensive datasets. Reasoning, according to interviewed researchers, is crucial for AI to achieve human or super-human intelligence levels. Current AI models excel at summarizing texts and composing prose but often struggle with common-sense problems, logical fallacies, and tasks like playing tic-tac-toe, leading to “hallucinated” information.
Researchers agree that improving reasoning in AI models is key to enabling AI to make major scientific discoveries and develop new software applications. OpenAI CEO Sam Altman emphasized the importance of progress in AI reasoning earlier this year.
Strawberry’s Role in OpenAI’s Strategy
Strawberry is central to OpenAI’s strategy to overcome reasoning challenges. The internal document detailed Strawberry’s goals but not the specifics of its implementation. Recently, OpenAI has hinted to developers and other stakeholders that it is close to releasing technology with significantly advanced reasoning capabilities.
OpenAI is not alone in this endeavor. Companies like Google, Meta, and Microsoft, along with academic labs, are experimenting with various techniques to improve AI reasoning. However, opinions differ on whether large language models can incorporate long-term planning and human-like reasoning. Notably, AI pioneer Yann LeCun of Meta has expressed skepticism about LLMs achieving human-like reasoning.
As OpenAI continues to develop Strawberry, the AI community watches closely, anticipating potential breakthroughs that could redefine the future of artificial intelligence.
Related topics:
How to use OpenAI for commercial applications
3 OpenAI’s Competitors: Leading Players in the AI Landscape