The Association for the Advancement of AI (AAAI), a venerable scientific outfit founded in 1979, recently had 25 of its researchers survey 475 members of the AI community about where things are headed. The result was a collective eye-roll at the tech industry's favourite pastime of scaling up. 

A full 76 per cent of respondents said it's “unlikely” or “very unlikely” that today’s brute-force approach—throwing ever-larger server farms and increasingly ridiculous GPU clusters at the problem—will get us to artificial general intelligence (AGI).

UC Berkeley computer boffin Stuart Russell, who helped write the report, told New Scientist:  “The vast investments in scaling, unaccompanied by any comparable efforts to understand what was going on, always seemed to me to be misplaced... about a year ago, it started to become obvious to everyone that the benefits of scaling in the conventional sense had plateaued.”

Essentially, he warned that everything has reached a limit, and it is time to stop building larger models and start conducting actual scientific research.

In other words, Silicon Valley might finally have to stop believing that AGI will magically emerge if they train a model the size of Greenland.

The tech giants, of course, remain in denial. Last December, Google CEO Sundar Pichai cheerily announced that the era of easy AI gains was over—then, without pausing for breath, insisted that the industry could “keep scaling up.”

Even OpenAI, the cathedral of the scaling doctrine, seems to be having second thoughts. After realising its next-gen GPT model wasn’t exactly setting the world on fire, they started dabbling in *test-time compute*—basically, letting the AI take longer to think before blurting out nonsense.

It delivered a performance boost without requiring a nuclear-powered data centre, but as Princeton's Arvind Narayanan warned, "don’t get too excited: it’s unlikely to be a silver bullet."