Photo Credit: Reuters
OpenAI is developing an innovative AI project codenamed “Strawberry,” aimed at enhancing the reasoning abilities of its AI models, according to sources familiar with the matter and internal documents reviewed by Reuters.
The project, previously undisclosed, is part of OpenAI’s efforts to demonstrate that its AI models can achieve advanced reasoning capabilities. Teams within OpenAI are actively working on Strawberry, as outlined in a recent internal document seen by Reuters in May. The exact timeline for public release of Strawberry remains unclear.
Details about how Strawberry operates are closely guarded within OpenAI. According to sources, the project aims to enable OpenAI’s AI to not only generate answers but also to plan autonomously and effectively navigate the internet for what OpenAI terms “deep research.”
This capability has been a challenge for AI models thus far, as noted by numerous AI researchers. OpenAI, in response to queries about Strawberry, emphasized its ongoing commitment to enhancing AI models’ understanding of the world, though specific questions regarding Strawberry were not directly addressed.
Previously known as Q*, the Strawberry project has been viewed within OpenAI as a significant breakthrough. Demonstrations earlier this year reportedly showcased AI’s ability to tackle complex scientific and mathematical questions beyond current commercial models’ reach.
Improving reasoning skills in AI models is seen as pivotal for achieving human-like or super-human intelligence. While existing large language models excel in tasks like text summarization and composition, they often struggle with common-sense reasoning problems. Enhancing these capabilities could potentially enable AI to make major scientific discoveries and develop sophisticated software applications.
OpenAI CEO Sam Altman has highlighted reasoning ability as crucial for AI advancement. Other tech giants like Google, Meta, and Microsoft are also exploring various approaches to enhance AI reasoning, reflecting a broader industry focus on this area of research.
Strawberry represents a critical component of OpenAI’s strategy to address these challenges. It involves specialized methods to refine AI models after initial training, potentially leading to significant advancements in their performance and capabilities.
In recent communications, OpenAI has signaled to developers and other stakeholders its impending release of technology featuring advanced reasoning capabilities. This includes adapting base models through post-training techniques like fine-tuning, aimed at enhancing specific aspects of model performance.
Strawberry’s approach shares similarities with Stanford’s “Self-Taught Reasoner” (STaR) method, which allows AI models to iteratively improve and potentially achieve intelligence levels surpassing human capabilities.
Ultimately, OpenAI aims for Strawberry to enable AI models to undertake long-horizon tasks, such as planning and executing complex actions over extended periods. This includes autonomous web browsing for research purposes and potentially automating tasks in software engineering and machine learning.
The development of Strawberry underscores OpenAI’s commitment to pushing the boundaries of AI reasoning, marking a significant step toward more capable and versatile AI systems.
