This is a special post for quick takes by DragonGod. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
I don’t know! Any estimate I can give here feels pretty fake, as in, I expect that my hypothesis space is too impoverished for such estimates to be meaningful. I also haven’t thought about it that much, since I’d need to become pretty confident (over 90%) in short timelines for me to want to stop my other research bets, and I don’t think I could gain that much confidence if I tried.
My own intuitions suggest that building ASI is probably hard (i.e. on the level of effort of one or a few Claude Shannon careers). I think it’s likely that we’re more bottlenecked by concepts than compute, I think those concepts are “far” in the sense that they require insight to be discovered, and it does not seem to me that people are on the right track for figuring out those concepts (including e.g. via modern AIs trying to improve current systems). Ideally I would say why I think these things, but it’s too much to get into in a short space.
There is currently a lot of effort going toward creating ASI, based on an approach which seems like it could plausibly get us to ASI, so my distribution is kind of bimodal. I think there’s a decent chance (maybe 20-30%?) we create superintelligence this decade. If this paradigm caps out, then I’m back to having roughly uniform, albeit skewed toward earlier, timelines. Uniform because it seems hard, absent knowing more about what the projects are, to pin down the timing of conceptual insights. Skewed earlier because although I don’t think the relationship between compute and developing intelligence is straightforward, it does seem like we'll have a ludicrous amount of it soon and, all else equal, that should make it easier to develop ASI (e.g. via emulating all of evolution or other crazy schemes).
I’m not sure where that leaves me, exactly. Mostly I want to say something like “I just don’t really know.” But 2040-2050 medians sound reasonable to me.