What is OpenAI’s Strawberry? AI Model Deep Research and Reasoning

ChatGPT maker OpenAI is reportedly building a new AI model called ‘Strawberry’ to help its artificial intelligence tools reach human-level intelligence through advanced reasoning. According to an OpenAI representative who spoke with Reuters, this could entail anything from producing ground-breaking scientific findings to creating brand-new software applications. Read this article to learn all about OpenAI’s secret project aimed at delivering advanced reasoning capabilities. 

What is Strawberry by OpenAI?

The OpenAI project Strawberry was formerly known as Q. It is a key component of OpenAI’s plan to overcome various challenges. The project includes a specialised way of “post-training” OpenAI’s generative AI models, or adapting the base models to hone their performance in specific ways after they have already been “trained” on reams of generalised data, one of the sources said.

Strawberry would be able to solve more complicated real-world problems at scale by efficiently scanning the internet on its own and reliably carrying out what OpenAI refers to as “deep research,” according to a July 12 report by Reuters.

Reuters further states that Strawberry bears resemblance to a Stanford technique named “Self-Taught Reasoner,” or “STaR,” which was developed in 2022. One of STaR’s creators, Stanford professor Noah Goodman, told Reuters that the technology allows AI models to “bootstrap” themselves into higher intelligence levels by iteratively creating their training data. In theory, this could be used to get language models to surpass human intelligence.

What is AGI in AI? Check out the Complete Explanation here!

How does the OpenAI Strawberry work?

The exact workings of Strawberry are closely guarded.

Strawberry intends to not just generate answers to queries but to plan enough to navigate the internet autonomously and reliably to perform what OpenAI terms “deep research,” according to the source.

Among the capabilities Strawberry aims to perform long-horizon tasks (LHT), the document says, referring to complex tasks that require a model to plan and perform a series of actions over an extended period, the first source explained.

Meanwhile, Sam Altman’s company recently collaborated with Los Alamos National Laboratory, the organisation that created the first nuclear weapon ever, to test the application of AI in bioscience research.

What is the new update about OpenAI Strawberry?

OpenAI’s model, O1 (formerly code-named “Strawberry”), introduces a novel approach to AI reasoning, focussing on multi-step logical problem-solving rather than purely scaling up model size. This model can solve complex tasks that previous models like GPT-4 could not handle. It uses reinforcement learning to improve reasoning across various subjects, including math, coding, and sciences, achieving better performance in areas like the AIME exam. However, O1 is slower and lacks multimodal capabilities like GPT-4.

How to Use AI to Plan Your Vacation (Step-by-Step Guide)?