"Barbs on Berries"

The Strange and Big Case of Secret Tech *Q and "Strawberry"

Have written before here about Elon Musk’s Secret Technology known only as *Q.

This was a development at OpenAI; an effort to achieve the mythical AGI - Artificial General Intelligence.

Then a falling our between Elon and Sam Altman.

Then a litigation where *Q was front and center (albeit ‘veiled’). It was reportedly licensed already to Microsoft and was either complete or not complete (with accounts unclear and conflicting). In one astonishing line, it was admitted that the parties perhaps did not even fully know what *Q was (!).

Mystery. Intrigue.

That case was dismissed by Elon Musk. Yet a new case was filed by him. And afterwards OAI announced that the former *Q would now be called, “Strawberry.”

So I went to Perplexity.ai for some non-biased general data and found more cause for mystery.

From a legal standpoint, this Perplexity account looks about as messy as could be, with 'building upon prior' and also 'new initiative' -- its a legal straddle.

Either some arrangement for settlement was made, or Strawberry/Q will remain a big part of the current pending litigation.

OAI seems to be “full steam ahead.”

In Perplexity’s account, the mythical buzzword AGI seems to be taken off the table which is another data point.

Answer from Perplexity

OpenAI is developing a new AI model codenamed "Strawberry," which appears to be an evolution of their earlier Project Q* initiative.

Here's what we know about these projects based on the available information:

Project Q*
Project Q* was an earlier AI initiative by OpenAI that aimed to improve reasoning capabilities in AI models. It was reportedly seen as a breakthrough within the company, with demos capable of solving advanced mathematical and scientific problems.

Project Strawberry
Strawberry is OpenAI's latest project focusing on enhancing AI reasoning abilities:

Core Objectives:
Develop AI models that can perform advanced human-like reasoning
Enable "deep research" capabilities, allowing AI to navigate the internet autonomously
Improve AI's ability to plan ahead and perform complex, long-horizon tasks

Key Features:
Builds upon the earlier Project Q* initiative
Inspired by Stanford's Self-Taught Reasoner (STaR) approach
Focuses on autonomous training data generation
Utilizes a specialized "post-training" method to adapt base models

Potential Applications:
Solving complex math and science problems
Navigating online environments
Conducting research by browsing the web autonomously
Performing tasks typically done by software and machine learning engineers

Development Status:
Described as a "work in progress" with no known release date
Tightly kept secret, even within OpenAI

Performance Indicators:
An OpenAI model (possibly related to Strawberry) reportedly scored over 90% on a MATH dataset, a benchmark for championship math problems

While specific details about Strawberry's inner workings remain confidential, it represents OpenAI's efforts to push the boundaries of AI reasoning capabilities. The project aims to move beyond simple query responses to more advanced, autonomous problem-solving and research abilities

AI Counsel - appears this highly-advanced Generative AI technology looks to achieve "both sides reasoning" (i.e. not just the best competent answers, but a system which would generate itself the most competent inputs matched for the most competent outputs = AUTOMATED BILATERAL AI). Much more to be seen from this legal and AI adventure…

“Automated Bilateral Artificial Intelligence”