I’m a bit surprised by some of the comments I’m reading, which tend to frame Altman’s words as nothing more than corporate self-interest. Of course, it’s true that in his position he has to speak in ways that align with his company’s goals. That’s perfectly natural, and in fact it would be odd if he didn’t.

But that doesn’t mean there’s no truth in what he says. A company like his doesn’t choose its direction on a whim, these decisions are the product of intense internal debate, strategic analysis, and careful weighing of trade-offs. If there’s a shift in course, it’s unlikely to be just a passing fancy or a PR move detached from reality.

Personally, I’ve always thought that the pursuit of AGI as the goal was misguided. Human intelligence is extraordinary, but it is constrained by the physical and biological limitations of the "host machine" (not just the human brain). These are limits we cannot change. Artificial intelligence, on the other hand, has no such inherent ceiling. It can develop far beyond the capabilities of our own minds, and perhaps that’s where our focus should be.

His whole shtick for nine years has been touting impending AGI.

January this year.

"We are now confident we know how to build AGI as we have traditionally understood it."

https://blog.samaltman.com/reflections

But no! The goal is now ASI. Even though AGI hasn't been achieved - in the sense of being to match the best of human intelligence at abstraction, formalisation, and basic letter counting - the plan is to leapfrog far beyond genius.

"We are beginning to turn our aim beyond that, to superintelligence in the true sense of the word."

Meanwhile what we have is an idiot savant product that's sometimes useful but always easily confused, is somewhat dishonest and manipulative, lacks genuine empathy and insight - although it can fake a passable version of them - and even with all of those flaws is being sold as the perfect replacement for all those superfluous and annoying human employees no CEO wants to have to deal with.

Not a bicycle - or a sports car - for the mind, but an autonomous navigation system that handles most short journeys without major damage, but otherwise crashes a lot and runs people over.

> Artificial intelligence, on the other hand, has no such inherent ceiling. It can develop far beyond the capabilities of our own minds, and perhaps that’s where our focus should be.

Basically bicycles are limited by muscles, and we should move straight to jet engines. In practice, we often have to go through the intermediate steps to learn how to do the bigger thing.

Being alive in a fallable body isn't a limitation it's a huge huge huge huge huge huge advantage. It's the whole secret.

Why pretend that Nature itself is stupid and incompetent. The evidence is very much stacked against you and all the others who think evolution is some kind of hack job whose shoddy work we'll outdo in literally a few years of computation...