Reuters reports a new project, code-named "Strawberry," which some on X speculate could be the next iteration of the infamous "reasoning mode" Q* leaked last year. Project Strawberry appears to be a new model or system with enhanced reasoning capabilities, including the ability to go online to gather information to solve particularly complex problems.
What do we know about Project Strawberry?
According to the source, Strawberry is a closely guarded secret within OpenAI. Although it might eventually integrate with ChatGPT, its primary purpose is to conduct "deep research" beyond handling basic user queries.
One of its capabilities includes the ability to autonomously and reliably navigate the internet while planning responses, without user prompts.
ChatGPT currently has the ability to search the web for specific details, but this feature is not always reliable. Often, users must explicitly instruct it to seek live data rather than relying solely on its training data.
OpenAI has not directly commented on the reports regarding Strawberry. However, the organization informed Reuters that it is standard practice to continuously research and develop a variety of new AI capabilities.
This ongoing exploration into advanced AI features, like those speculated for Project Strawberry, underscores OpenAI's commitment to pushing the boundaries of what artificial intelligence can achieve.
When you see project Strawberry
Project Strawberry is a research initiative by OpenAI, and it remains uncertain if it will ever be released as a standalone product. Nonetheless, the insights gained from this project are likely to inform future AI models and product updates developed by OpenAI.
I anticipate that any advancements in reasoning will be integrated into the training for GPT-5. Project Strawberry might emerge as a groundbreaking model, similar to Omni, which revolutionized the field unexpectedly. Unlike GPT-4, GPT-4o (Omni) was inherently multimodal, both in input and output.
Strawberry could represent a new class of models, designed as native reasoners from the outset, eliminating the need for fine-tuning or additional training. Although little is known about it, OpenAI hinted to Reuters that achieving human-like understanding in AI is a crucial objective.
