OpenAI, the company behind ChatGPT, is reportedly working on an innovative AI project known as "Strawberry."
According to sources familiar with the matter and internal documents reviewed by Reuters, this project, which was previously called Q, represents a significant breakthrough within the company.
Although OpenAI has not officially disclosed details about Strawberry, the startup is striving to demonstrate that its models can deliver advanced reasoning capabilities. OpenAI teams are actively working on Strawberry. However, the precise release date of this technology remains uncertain.
The specifics of how Strawberry functions are tightly guarded, even within OpenAI. The project aims to enhance the AI's ability not only to generate answers but also to autonomously navigate the internet for in-depth research. This capability has been a challenging goal for AI models to date, as noted by several AI researchers.
When asked about Strawberry, an OpenAI spokesperson stated, "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."
Earlier this year, demonstrations of Q reportedly showcased the model's ability to answer complex science and math questions beyond the reach of current commercially available models.
During an internal meeting on Tuesday, OpenAI allegedly presented a demo of a research project claiming new human-like reasoning skills.
OpenAI hopes that Strawberry will significantly enhance its AI models' reasoning abilities. This involves a specialized post-training process after the AI model has been pre-trained on extensive datasets. Researchers agree that improved reasoning is crucial for AI to achieve human or super-human intelligence.
While large language models can efficiently summarize dense texts and compose sophisticated prose, they often struggle with common sense problems that humans intuitively solve. Addressing these issues is essential for AI models to achieve advanced reasoning capabilities.
Strawberry is part of OpenAI's broader strategy to overcome these challenges. The company has hinted to developers and other parties that it is close to releasing technology with significantly advanced reasoning capabilities. Strawberry involves a specialized post-training method to refine the performance of generative AI models after their initial training on generalized data.
This project shares similarities with Stanford's 2022 method called "Self-Taught Reasoner" (STaR), which allows AI models to iteratively create their own training data to reach higher intelligence levels. One of its creators, Stanford professor Noah Goodman, noted the potential and implications of such advancements.
OpenAI aims to use Strawberry for performing long-horizon tasks (LHT), requiring the model to plan and execute a series of actions over extended periods. The company is developing and testing the models on a "deep-research" dataset, although specifics about the dataset and the duration of these tasks remain unknown.
Additionally, OpenAI plans for its models to autonomously conduct research online with the help of a "CUA" (computer-using agent) that can take actions based on its findings. The company also intends to test these capabilities in software and machine learning engineering tasks.