OpenAI’s Strawberry launching this fall?

First came reports that OpenAI was working on a secret project, codenamed Project Strawberry, which was a huge step towards achieving Artificial General Intelligence (AGI) which means AI can perform tasks at a human-level.

Next came a cryptic social media post—from CEO, Sam Altman—of a bunch of strawberries, seemingly confirming this report. 

Now, insiders at OpenAI are telling us that Project Strawberry will land imminently: This Fall.  

What is Project Strawberry? 

Previously codenamed Q*, Project Strawberry was the brainchild of former OpenAI chief scientist, Ilya Sutskever, who left the company earlier this year over concerns over security. 

Strawberry can reportedly understand and answer nuanced questions about more subjective topics, like marketing strategies, has advanced programming skills, and can solve math equations, without training, recently scoring 90% on Math benchmark tests - something which is unheard of in AI. 

It’s being hailed as the next step towards achieving AGI because it has advanced self-learning and logical thinking capabilities which means it can continue to evolve without human guidance and make reasoned decisions using available data. 

The larger version of Strawberry (the smaller version is likely to be released in the fall) can also generate high-quality ‘synthetic’ data which will reduce the need to rely on internet-based images and content and is likely to reduce hallucinations. 

If rumors are true, OpenAI’s newest ChatGPT model—Orion—is currently being trained on this data, and will be powered by Strawberry.