AI progress can be rapid but the pathway to it may involve different capability unlocks. For example, it may be you automate work more broadly and then reinvest that into more compute/automate chipmaking itself). Or you can get the same unlocks without rapid progress. For example, you get a superhuman coder but run into different bottlenecks.
I think it’s pretty obvious AI progress won’t completely stall out, so I don’t think that’s the prediction you’re making? It’s one thing to say AI progress won’t be rapid and then give a specific story as to why. Later if you hit most of your marks, it’ll look like a much more valuable prediction than saying simply it won’t be rapid. (Same applies to AI 2027).
The authors of AI 2027 made a pretty specific story before the release of ChatGPT and looked really prescient after the fact since it turned out to be mostly accurate.
AI progress can be rapid but the pathway to it may involve different capability unlocks. For example, it may be you automate work more broadly and then reinvest that into more compute/automate chipmaking itself). Or you can get the same unlocks without rapid progress. For example, you get a superhuman coder but run into different bottlenecks.
I think it’s pretty obvious AI progress won’t completely stall out, so I don’t think that’s the prediction you’re making? It’s one thing to say AI progress won’t be rapid and then give a specific story as to why. Later if you hit most of your marks, it’ll look like a much more valuable prediction than saying simply it won’t be rapid. (Same applies to AI 2027).
The authors of AI 2027 made a pretty specific story before the release of ChatGPT and looked really prescient after the fact since it turned out to be mostly accurate.