TLDR
- OpenAI is developing a new AI model called “Strawberry” to improve reasoning capabilities.
- Strawberry aims to perform “deep research” by autonomously navigating the internet.
- The project extends OpenAI’s previous Q* project announced in November 2023.
- Strawberry could potentially solve complex real-world problems at scale.
- The project is still in development, with details kept secret even within OpenAI.
OpenAI, the company behind ChatGPT, is working on a new artificial intelligence (AI) model called “Strawberry.” This project aims to improve the reasoning abilities of AI, allowing it to tackle more complex real-world problems.
Strawberry is designed to perform what OpenAI calls “deep research.” This means the AI could search the internet on its own to find information and solve problems.
It’s an upgrade from current AI models, which often struggle with tasks that require advanced thinking or common sense.
The project builds on OpenAI’s earlier work, known as Q*, which was announced in November 2023. Some experts called Q* a breakthrough that could lead to much more powerful AI models.
New:🍓OpenAI's secretive Q* project is now called Strawberry! 🍓 Read the story for insight into how it works and what OpenAI intends to use it for. Can't wait to test out its next-level reasoning skills myself! Yum🍓https://t.co/f7dY3kcfki
— Anna Tong (@annatonger) July 12, 2024
Strawberry takes this idea further, aiming to make AI think more like humans do.
One key feature of Strawberry is its ability to plan ahead when responding to questions. It could navigate the internet without needing specific instructions from users.
This is different from current versions of ChatGPT, which need to be told to look for up-to-date information.
OpenAI hopes Strawberry will be able to do things like make scientific discoveries or create new software. The company is even exploring the use of a “computer-using agent” that could act on the AI’s findings, pushing the boundaries of what AI can do on its own.
The development of Strawberry involves a special process applied to AI models after they’ve been trained on large amounts of data. This helps the AI produce more human-like responses.
OpenAI is testing these new models on a special “deep-research” dataset, though the details of this dataset are not public.
While current AI models are good at summarizing text and writing, they often struggle with tasks that humans find easy.
For example, they might have trouble identifying flawed logic or playing simple games like tic-tac-toe without making mistakes. Strawberry aims to improve on these areas.
OpenAI is keeping many details about Strawberry secret, even within the company. It’s not clear how close the project is to being finished or when it might be available to the public.
An OpenAI spokesperson said that researching new AI abilities is common in the industry, as companies believe these systems will get better at reasoning over time.
The development of Strawberry comes as OpenAI faces other changes. Recently, Microsoft and Apple reportedly stepped down from OpenAI’s board due to increasing regulatory scrutiny.
OpenAI has also partnered with Los Alamos National Laboratory to explore using AI in bioscience research.
