OpenAI, the company behind ChatGPT, is developing a secretive AI project code-named “Strawberry” designed to significantly enhance AI reasoning capabilities, potentially leading to breakthroughs in autonomous research and problem-solving.
While details remain tightly under wraps, Reuters has learned through sources and internal documents that Strawberry involves a novel approach to training and processing AI models, enabling them to perform tasks that have eluded existing systems.
Unlike current models that primarily focus on generating text-based responses, Strawberry aims to equip AI with the ability to “plan ahead” and navigate the internet autonomously to conduct what OpenAI calls “deep research.” This represents a significant leap forward, as it requires a deeper understanding of context, logic, and multi-step problem-solving.
The pursuit of human-level AI reasoning is a central focus across the industry, with companies like Google, Meta, and Microsoft all exploring various techniques. Experts believe that achieving this breakthrough could unlock the potential for AI to drive scientific discoveries, develop complex software, and tackle challenges that currently require human intuition and planning.
While OpenAI has not publicly confirmed specifics about Strawberry, a company spokesperson told Reuters, “We want our AI models to see and understand the world more like we do. Continuous research into new AI capabilities is a common practice in the industry, with a shared belief that these systems will improve in reasoning over time.”
Strawberry appears to be an evolution of an earlier OpenAI project known as Q*, which garnered internal excitement for its advanced reasoning abilities. Sources who witnessed Q* demos reported its ability to solve complex math and science problems beyond the capabilities of current commercially available AI.
While the exact mechanisms remain secret, sources suggest that Strawberry involves a specialised form of “post-training” — a process of refining AI models after they’ve been trained on massive datasets. This post-training phase, potentially involving techniques like “fine-tuning” and self-generated training data, is crucial for honing the AI’s reasoning abilities.