OpenAI, with backing from Microsoft, is reportedly developing a groundbreaking AI project known as “Strawberry,” according to an insider and internal documents reviewed by Reuters.

This innovative initiative aims to significantly enhance the reasoning abilities of AI models, an area where existing technology has struggled. Strawberry represents a critical effort by OpenAI to advance AI’s capacity for deep reasoning, potentially enabling models to autonomously navigate the internet and conduct extensive research.

Details about Strawberry remain closely guarded within OpenAI. The project seeks to develop AI that can not only generate responses but also plan and execute complex tasks independently, according to internal documents and sources familiar with the matter. This capability has long been a challenge for AI, as current models often fail at common-sense reasoning and complex problem-solving.

When asked about the project, an OpenAI spokesperson stated that continuous research into new AI capabilities is essential, with the aim of making AI understand and interact with the world more like humans do. However, the spokesperson did not provide specific details about Strawberry.

Previously known as Q*, Strawberry has been internally celebrated for its potential breakthroughs. Early demos of Q* reportedly showcased the ability to tackle complex science and math problems beyond the reach of current commercial models.

During a recent internal meeting, OpenAI demonstrated a project with new human-like reasoning abilities, although it was unclear if this was Strawberry. The company is focused on improving the reasoning capabilities of its models, which involves a specialized process of refining AI models post-training.

Experts agree that enhancing AI reasoning is crucial for achieving advanced, human-like intelligence. While current models excel at generating text, they often struggle with intuitive problem-solving and logical consistency.

OpenAI’s CEO, Sam Altman, has emphasized the importance of reasoning in AI development. Strawberry is part of OpenAI’s strategy to address these challenges by implementing advanced post-training techniques, akin to methods developed at Stanford, such as the “Self-Taught Reasoner” (STaR).

Strawberry aims to handle long-term tasks that require planning and a series of actions over time. OpenAI is creating a “deep-research” dataset to train and evaluate these models, though specifics about the dataset remain undisclosed.

Ultimately, OpenAI envisions its models using these advanced reasoning capabilities to conduct autonomous research online and assist in software and machine learning engineering tasks.

It would be very interesting to see how this model will be utilized by developers to solve real life problems in India.

Pic Credit: OpenAI Wiki page