Yeah... I seem to remember seeing this before, only it was by the end of 2027. So the schedule is slipping by one year per year. (In fairness, it might not have been Altman who made the previous prediction.)
But one year slip per year is not the pattern of a successful project - it's the pattern of a floundering one. You see this sometimes in projects where they still haven't figured out what the spec is for what they're trying to build. (So, do they know the spec for building a superintelligence? I'm pretty sure that no, they don't.)
What they have is evidence that they're making progress, and a completely-without-evidence idea of how much further ahead superintelligence might be, and an extrapolation based on progress continuing at the same rate. Well, the part that is the most suspect is the guess as to how far away superintelligence is. If that's wrong, the whole estimate is worthless.
I think what's amazing to me is, we used to have Steve Jobs' exaggerations, the "reality distortion field" and correct me if wrong, but he basically delivered on the visions he had. Then Musk started to ratchet up the lies more and more, but I think Trump coming into the office essentially supercharged this idea in Silicon Valley that actually, lying was more profitable than truthtelling.
As someone who's trusting, I do sort of listen to Altman or Amodei (who I think has been a bit more truthful in his predictions, a year from AI writing all software actually ended up being more truthful than I think people though, even if it isn't technically true) and like have this nagging voice in the back of my head that these people know something I don't, but then, just looking at this clear leadership trend seeming to suggest that lying is more profitable than trying to tell the truth, and the whole picture definitely does not look clear.
Full tweet (since it wouldn't fit in the title): Sam Altman: Superintelligence probably by end of 2028. So we got roughly 2 years left. Enjoy your job while you still can. Time is ticking.
But one year slip per year is not the pattern of a successful project - it's the pattern of a floundering one. You see this sometimes in projects where they still haven't figured out what the spec is for what they're trying to build. (So, do they know the spec for building a superintelligence? I'm pretty sure that no, they don't.)
What they have is evidence that they're making progress, and a completely-without-evidence idea of how much further ahead superintelligence might be, and an extrapolation based on progress continuing at the same rate. Well, the part that is the most suspect is the guess as to how far away superintelligence is. If that's wrong, the whole estimate is worthless.