What is "Q*", which has suddenly appeared in the OpenAI-Altman controversy?

//Summary - Level-C2//

The sudden firing and rehiring of OpenAI CEO Sam Altman are linked to a top-secret AI program called Q* (Q Star), causing internal conflict over its potential threat to humanity. Altman's dismissal, reportedly due to prioritising development over safety concerns, led to a swift employee revolt. Speculation surrounds Q*'s capabilities, suggesting it may surpass AGI in solving complex problems. External experts remain sceptical, while AI enthusiasts anticipate transformative advancements. The drama has sparked excitement and uncertainty, prompting varied interpretations.

A)
"There are moments when you're scared to death, wondering if what you've created is a tool or a monster."
(Sam Altman, on the eve of his CEO transition, at the Robot Heart event in Oakland)

The morning after making such a powerful statement, Sam Altman was suddenly fired as CEO of OpenAI.

Microsoft, his largest shareholder, immediately hired him, saying: "Come to me!" However, 710 of OpenAI's 770 employees made a joint appeal, saying, "If you don't let him come back to work, we'll quit too," he was successfully reinstated as CEO. All the executives who carried out the coup were fired, except for one who was also Quora's CEO, and the reshuffle ended after four days. However, the mystery of why he was forced to quit in the first place remains unsolved.

The key to solving this mystery is said to be the above statement, which CEO Altman blurted out on stage at Robot Heart, held in an Oakland warehouse on the eve of his dismissal. Below is the full version.

AI AND THE FUTURE OF ART - SAM ALTMAN AND ANDROID JONES - PLAYALCHEMIST ROBOT HEART RESIDENCY
https://www.youtube.com/watch?v=NFoAR1S5hEE

B)
Six months ago, we said clearly, "What we're making are tools, not monsters." But...

AI is a tool, not a creature, says Sam Altman, CEO of OpenAI
https://youtu.be/EdM_1jaeL_w

What has suddenly happened?
It's a drama about a change of heart. There has to be something behind it! When Reuters sniffed around, an anonymous source inside the company said: "It seems there has been a breakthrough that will shake humanity's very existence, and even the board is spooked."

C)
The breakthrough was a top-secret development programme called Q* (Q Star), known only to a handful of people within the company. The background to the sudden dismissal was that in-house AI researchers reported to the board of directors about the discovery of a powerful AI that could pose a threat to humanity, leading to a direct conflict between executives who wanted to prioritise safety and the CEO who tried to push development.

It's hard to believe, and it seems a little impossible, that the world's brain trust, OpenAI, would overreact to a single programme like this. It is said that if this "Q*" is passed, an AI waiting for instructions can solve primary school-level mathematics on its own, and this seems to be a feat that breaks the barrier of "AGI *". A similar story was told in an interview with The Information.

OpenAI Made an AI Breakthrough Before Altman Firing, Stoking Excitement and Concern
https://www.theinformation.com/articles/openai-made-an-ai-breakthrough-before-altman-firing-stoking-excitement-and-concern

*AGI (Artificial General Intelligence): AI with a wide range of knowledge can decide "what to do" independently, like a human.

OpenAI has neither confirmed nor denied this, so various rumours and speculations are flying around. Finally, someone said, "Maybe Q is the Q of Q-Learning?"

D)
What is Q-learning?
AI has three approaches to learning:

No.1: Supervised learning: A learning method feeds data with learning labels attached. Example) OpenAI's ChatGPT.

No.2: Unsupervised learning: A learning method in which AI uses an algorithm to find patterns in a pile of data without learning labels. For example, Spotify and Netflix have a feature that recommends new content based on past viewing patterns.

No.3: Reinforced Learning (RL): A learning method that encourages goal attainment by providing rewards. Q-learning is a type of RL. It's like training a dog by feeding it. Actions that produce good results are rewarded, and those with negative results are punished. The AI is then 'trained' to find a way to maximise results in a given situation.

Q-learning involves repeated trial and error to find the best way to achieve a goal, which appears to have been achieved using mathematical formulae.

E)
Experts are sceptical
Immediately, voices from outside the company say this is just a breakthrough in AGI in general. MIT Technology Review quotes Wenda Li, an AI lecturer at the University of Edinburgh.

"Researchers have been working on solving mathematical problems with AI models for years. Large-scale language models such as ChatGPT and CPT-4 can solve maths problems but are not very good at it and can be hit or miss. No algorithm or architecture can consistently solve mathematical problems using AI."

"Deep learning and transformers (a type of neural network) used in language models are good at pattern recognition, but that's not enough."

F)
AI watchers are excited
Still, AI watchers are excited by the internal drama, which is like a rematch of Jobs' firing compressed into Friday, Saturday, Sunday and Monday. Looking at the summary of public observations (above), I found: "If human feedback is included in learning, the quality will be poor, and not good, but we can get AI to do that part (AI scores AI)". I was half excited, half scared, with some saying, "If that happens, we could see a transcendental evolution like Alpha GO".

I feel the truth is that "people are just making noise and making things louder", and Altman is working on an AI device to replace the iPhone, along with Jony Ive (whose contract with Apple ended last year and who joined Yugun). They also show signs of teaming up with Masayoshi Son, a Japanese high-technology company CEO, to form a separate company.

There may have been some executives who were not happy about this, but there was something that scared the board, too.

What is "Q*", which has suddenly appeared in the OpenAI-Altman controversy?

https://www.gizmodo.jp/2023/12/sam-altman-openai-q-machine-learning-artificial.html

OpenAI announces the return of Sam Altman as CEO and the new structure

https://www.watch.impress.co.jp/docs/news/1550918.html

OpenAI announced on the 29th (US time) the return of Sam Altman as CEO and a new structure.

On the 17th, OpenAI's board of directors fired Altman as CEO. After that, there were signs that Mr. Altman would return, and on the 19th, it was announced that Mr. Altman would be joining Microsoft, but on the 22nd, Mr. Altman announced that he would return to OpenAI as CEO. This time, we told the new system.