OpenAI is developing a revolutionary reasoning technology in artificial intelligence
The leading artificial intelligence company, OpenAI, appears to be working on a new technology. According to reports, the Microsoft-backed company is developing a new reasoning technology.
The Microsoft-backed developer of ChatGPT, Sora, and DALL-E, OpenAI, is reportedly working on a mysterious project codenamed “Strawberry.” It is said that the company aims to introduce a new approach to the development processes of AI models with this project. The Strawberry project is expected to add advanced reasoning capabilities to OpenAI’s large language models. According to a report by Reuters, OpenAI is keeping plans and details about how Strawberry works secret, even within the company. Additionally, it is currently unclear when Strawberry will be publicly announced.
AI Planning for the Future
Moreover, thanks to Strawberry, AI will not only generate responses to queries but also plan ahead to autonomously and reliably navigate the internet through a mechanism the company refers to as “deep research.” There is also another significant detail: the Strawberry project is believed to be the mysterious Q* (pronounced Q-Star) project we have heard about in the past. A few months ago, it was suggested that Q* could be a breakthrough in the pursuit of artificial general intelligence (AGI), also known as superintelligence.
Additionally, last Tuesday, OpenAI presented a research project during an internal meeting, claiming it possesses new human-like reasoning abilities. It is unclear if this is the Strawberry project, but indications point in that direction.
Reasoning is one of the weakest points of modern AI. While large language models (the systems powering tools like ChatGPT) can summarize dense texts and generate prose, they fall short on topics requiring intuition, common sense, or thoughtful consideration. When faced with such questions, the models often produce fabricated information, a phenomenon known as “hallucination.”
AI systems strong in reasoning could excel in planning ahead, understanding the physical world, and solving complex multi-step problems. Enhancing reasoning in AI models could revolutionize their capabilities, enabling them to achieve everything from making significant scientific discoveries to planning and creating new software applications. OpenAI CEO Sam Altman mentioned in a conversation with Bill Gates in January that the “most important areas of advancement in AI will be around reasoning capabilities.”
Additionally, it is said that what OpenAI is working on bears similarities to a method developed at Stanford in 2022, known as “Self-Taught Reasoner” or “STaR.” STaR is an approach that allows AI models to iteratively create their own training data, elevating themselves to higher levels of intelligence. Theoretically, this approach could be used to enable language models to surpass human-level intelligence.
