OpenAI has been the center of attention in the artificial-intelligence world due to the dramatic firing and swift return of CEO Sam Altman. However, there has been speculation it was triggered by a significant breakthrough in AI might – but that might have been overblown.
Firstly some context – Reuters reported last week that staff researchers at OpenAI wrote a letter to the board warning an internal project named Q*, or Q-Star, could represent a breakthrough in creating an artificial general intelligence –an AI which can surpass humans in a range of fields– ahead of Altman’s firing.
OpenAI of course has already made a number of strides in AI. However, the suggestion that one particular advance might have played a role in Altman’s original firing seemed to raise the stakes. But the thought that we are on the brink of meeting genuinely intelligent machines, and all the risks that involves, might be too hasty.
Firstly it’s worth pointing out that neither OpenAI nor its largest backer
Microsoft
(ticker: MSFT) have publicly confirmed the existence of Q*, much less the possibility that it is a dangerous breakthrough in AI technology. OpenAI didn’t respond to requests for comment on the Reuters report.
As we noted last week, when a Google engineer claimed in 2022 that an unreleased AI system had become sentient, it caused a brief flurry of excitement before he was fired and the company denied the claim.
The only detail given in the report about Q*’s capabilities was that it could solve certain mathematical problems at the level of grade-school students. That has led to skepticism about how serious an advance it could be – including from Elon Musk, who jokingly suggested his own Grok chatbot will outdo it by both solving math problems and fundamental philosophical questions.
Still, that hasn’t stopped there being plenty of speculation about what Q* might really represent. That has mostly centered on the theory that it is a combination of two types of algorithm. The first is Q-learning, an algorithm for reinforcement learning, or trial-and-error methods. The second is A*, an algorithm that improves the efficiency of a search for a solution by effectively looking ahead to potential obstacles, which can minimize computation costs.
Combining search and learning technologies has been the source of significant AI advances before. Jim Fan, a research scientist at
Nvidia
(NVDA) noted that such a combination was what enabled the development of AlphaGo, the AI which became the best player in the world at the Chinese game of Go in 2017, shocking the technology and board-game worlds.
However, it’s a big leap to go from that to suggesting OpenAI is close to achieving a genuine step change in AI.
“Nothing says Q* will be more creative in writing poetry, telling jokes, or role playing. Improving creativity is a fundamentally human thing, so I believe natural data will still outperform synthetic ones,” Fan said on social-media platform X, formerly called Twitter.
Write to Adam Clark at [email protected]
Read the full article here