OpenAI is working on a new AI project called “Strawberry.”
This project aims to create an AI that can plan ahead, browse the internet by itself, and carry out detailed research.
The exact workings of Strawberry are a closely guarded secret at OpenAI.
The project involves a special method for improving AI models after they have been initially trained on large datasets.
Strawberry aims to enhance the AI’s ability to generate accurate answers, plan effectively, and conduct research on its own.
The goal is to overcome current AI limitations like issues with common sense and logical errors. Strawberry was once known as Q*, which was considered a significant breakthrough in AI reasoning.
Other tech giants like Google, Meta, and Microsoft are also working on improving AI reasoning.
There is ongoing debate among researchers about whether large language models can achieve long-term planning and human-like reasoning.
Strawberry includes a “post-training” phase, similar to the “Self-Taught Reasoner” (STaR) method from Stanford. It aims to handle complex tasks over a long period and utilize a “deep-research” dataset.
OpenAI hopes that Strawberry will greatly improve AI reasoning, enabling AI to conduct advanced research and perform software engineering tasks independently.
