Q* Confirmed As Real

Q* may have involved advances in the use of synthetic data to turbocharge LLM training.

A breakthrough in AGI, or advances in training existing LLMs?

Sam Altman, back in his role as OpenAI CEO, has confirmed that Q*—the internal project that, in part, precipitated the former board's decision to fire him—is real.

In an interview with The Verge, Altman was asked about the project, which has been rumored to show unexpected advances towards Artificial General Intelligence.

AGI: No Longer Sci-Fi?

In the interview, Altman declined to say why exactly the board fired him, something about which there has been much speculation. There is currently an independent review of the decision, and Altman says he is not ready to talk about it yet.

Critics have suggested that as well as there being tensions between the non-profit and for-profit arms of the company, rapid progress towards AGI might have taken the board by surprise, causing them to slam the brakes on by dismissing Altman.

In response to a different question, he answered, "And we’re making such great progress on the mission I care so much about, the mission of safe and beneficial AGI."

Little is publicly known about Q*, though it appears that the system displayed properties of AGI far in advance of what was previously thought possible.

Reuters cited an unnamed source, saying "Given vast computing resources, the new model was able to solve certain mathematical problems... Though only performing math on the level of grade-school students, acing such tests made researchers very optimistic about Q*'s future success." The Information claimed that Q* was viewed as a breakthrough that would lead to "far more powerful artificial intelligence models," and that "the pace of development alarmed some researchers focused on AI safety".

There is further intrigue in the news that the project was led by Ilya Sutskever, OpenAI’s chief scientist and cofounder, who actually participated in the board's coup but later expressed remorse.

Some researchers believe that Q* is simply a project to improve Large Language Models' mathematical abilities, something to which they are currently not well suited. Others have suggested that Q* used advances in so-called reinforcement learning, which uses positive and negative reinforcement to improve outcomes, and "synthetic data" generated by the AI, rather than relying on data from the web. "Synthetic data represents the shortest path to expanding datasets. In the short term, it’s clear we can create some useful data with this. What is not clear is the extent to which it can be scaled — ie can it replace internet scale data entirely?"

"Three Years Away"

Elon Musk recently estimated that we are very close to AGI being a reality. In the same DealBook Summit interview that saw him rage against advertisers who had pulled ads from Twitter/X, he commented, "If you say smarter than the smartest human at anything, it can write as good a novel as JK Rowling or discover new physics or invent new technology... I would say we are less than three years from that point."

Subscribe to our newsletter and follow us on X/Twitter.

Great! You’ve successfully signed up.

Welcome back! You've successfully signed in.

You've successfully subscribed to REX Wire.

Success! Check your email for magic link to sign-in.

Success! Your billing info has been updated.

Your billing was not updated.