Ever since last week’s dramatic events at OpenAI, the rumor mill has been in overdrive about why the corporate’s chief scientific officer, Ilya Sutskever, and its board decided to oust CEO Sam Altman.

While we still don’t know all the small print, there have been reports that researchers at OpenAI had made a “breakthrough” in AI that had alarmed staff members. Reuters and The Information each report that researchers had provide you with a brand new technique to make powerful AI systems and had created a brand new model, called Q* (pronounced Q star), that was in a position to perform grade-school-level math. In response to the individuals who spoke to Reuters, some at OpenAI imagine this may very well be a milestone in the corporate’s quest to construct artificial general intelligence, a much-hyped concept referring to an AI system that’s smarter than humans. The corporate declined to comment on Q*. 

Social media is filled with speculation and excessive hype, so I called some experts to learn the way big a deal any breakthrough in math and AI would be.

Researchers have for years tried to get AI models to unravel math problems. Language models like ChatGPT and GPT-4 can do some math, but not thoroughly or reliably. We currently don’t have the algorithms and even the best architectures to have the opportunity to unravel math problems reliably using AI, says Wenda Li, an AI lecturer on the University of Edinburgh. Deep learning and transformers (a sort of neural network), which is what language models use, are excellent at recognizing patterns, but that alone is probably going not enough, Li adds. 

Math is a benchmark for reasoning, Li says. A machine that’s in a position to reason about mathematics, could, in theory, have the opportunity to learn to do other tasks that construct on existing information, equivalent to writing computer code or drawing conclusions from a news article. Math is a very hard challenge since it requires AI models to have the capability to reason and to essentially understand what they’re coping with. 

A generative AI system that might reliably do math would wish to have a very firm grasp on concrete definitions of particular concepts that may get very abstract. Quite a lot of math problems also require some level of planning over multiple steps, says Katie Collins, a PhD researcher on the University of Cambridge, who focuses on math and AI. Indeed, Yann LeCun, chief AI scientist at Meta, posted on X and LinkedIn over the weekend that he thinks Q* is more likely to be “OpenAI attempts at planning.”

Individuals who worry about whether AI poses an existential risk to humans, one among OpenAI’s founding concerns, fear that such capabilities might result in rogue AI. Safety concerns might arise if such AI systems are allowed to set their very own goals and begin to interface with an actual physical or digital world in some ways, says Collins. 

But while math capability might take us a step closer to more powerful AI systems, solving these forms of math problems doesn’t signal the birth of a superintelligence. 

“I don’t think it immediately gets us to AGI or scary situations,” says Collins.  It’s also very necessary to underline what sort of math problems AI is solving, she adds.

“Solving elementary-school math problems could be very, very different from pushing the boundaries of mathematics at the extent of something a Fields medalist can do,” says Collins, referring to a top prize in mathematics.  

Machine-learning research has focused on solving elementary-school problems, but state-of-the-art AI systems haven’t fully cracked this challenge yet. Some AI models fail on really simple arithmetic problems, but then they’ll excel at really hard problems, Collins says. OpenAI has, for instance, developed dedicated tools that may solve difficult problems posed in competitions for top math students in highschool, but these systems outperform humans only occasionally.  

Nevertheless, constructing an AI system that may solve math equations is a cool development, if that’s indeed what Q* can do. A deeper understanding of mathematics could open up applications to assist scientific research and engineering, for instance. The flexibility to generate mathematical responses could help us develop higher personalized tutoring, or help mathematicians do algebra faster or solve more complicated problems. 

This can also be not the primary time a brand new model has sparked AGI hype. Just last 12 months, tech folks were saying the identical things about Google DeepMind’s Gato, a “generalist” AI model that may play Atari video games, caption images, chat, and stack blocks with an actual robot arm. Back then, some AI researchers claimed that DeepMind was “on the verge” of AGI due to Gato’s ability to achieve this many alternative things pretty much. Same hype machine, different AI lab. 

And while it is likely to be great PR, these hype cycles do more harm than good for your entire field by distracting people from the true, tangible problems around AI. Rumors about a strong recent AI model may additionally be an enormous own goal for the regulation-averse tech sector. The EU, for instance, could be very near finalizing its sweeping AI Act. Considered one of the largest fights right away amongst lawmakers is whether or not to present tech firms more power to control cutting-edge AI models on their very own. 

OpenAI’s board was designed as the corporate’s internal kill switch and governance mechanism to stop the launch of harmful technologies. The past week’s boardroom drama has shown that the underside line will all the time prevail at these firms. It’ll also make it harder to make a case for why they must be trusted with self-regulation. Lawmakers, take note.