Affiliate links on Android Authority may earn us a commission. Learn more.
OpenAI developing secret "Strawberry" project to help ChatGPT outthink humans
![OpenAI on website on smartphone stock photo (3) OpenAI on website on smartphone stock photo (3)](https://www.androidauthority.com/wp-content/uploads/2023/09/OpenAI-on-website-on-smartphone-stock-photo-3.jpg)
- OpenAI is working on a new AI project to achieve human-like reasoning capabilities.
- Dubbed Strawberry, the project aims to help AI achieve reasoning capabilites on par with humans.
As OpenAI’s next-generation GPT-5 language model inches closer to release, we’re finally getting our first glimpses at the potential AI breakthroughs in store. According to a new Reuters report, the Microsoft-backed startup is working on a new technique to improve AI reasoning capabilities under the code name “Strawberry.” This news comes from a source familiar with the project and internal documents reviewed by the publication.
We’ve known that OpenAI has been racing to develop artificial general intelligence (AGI) since late last year when news of a secretive Q* model leaked. Around the same time, multiple OpenAI employees wrote a letter to the board warning of a new AI breakthrough that “could threaten humanity.” It’s believed that Q* has since been renamed to Strawberry.
According to Reuters, OpenAI’s Strawberry project involves training the AI to “plan ahead enough to navigate the internet autonomously” and undertake “deep research.” The latter could help dispel hallucinations as language models typically resort to making up information when they lack training data on a particular subject.
Human-like reasoning would help eliminate hallucinations in AI responses.
A recent Bloomberg report also lends credence to Strawberry’s existence, stating that OpenAI demonstrated a research project capable of achieving “human-like reasoning” in an employee meeting last week. While little is known about Strawberry’s inner workings, it may bear similarities to a method known as Self-Taught Reasoner (STaR) that allows a model to improve itself by learning from its own reasoning. This means STaR could help AI models achieve better-than-human intelligence over the long run.
OpenAI is not the only artificial intelligence startup chasing advanced, human-like reasoning. Earlier this month, Perplexity upgraded its Pro Search feature to break down and resolve complex queries via multi-step reasoning to emulate human research and problem-solving.
However, OpenAI appears to be aiming higher with Strawberry, as it plans to develop AI that can browse the web independently and, presumably, perform actions like a human would. At this time, it’s unclear if Strawberry’s advanced reasoning capabilities will debut with GPT-4o or wait until GPT-5’s release.