Anthropic Shares Balanced Views On AI Economic Impacts

http://x.com/i/article/2079673806948290561

Good post with sensible takes from Anthropic on econ impacts!

This is absolutely excellent and gets at something that I find increasingly annoying. There are just many bottlenecks besides intelligence and they won't magically be solved by AGI. Even if you think that AGI will be able to deal with legal and political obstacles easily and quickly, a view I find hopelessly naive, even more fundamentally there are physical constraints that won't be abolished by AGI or even ASI, although it will allow us to work around them much more effectively. Concretely, even if after we have AGI, a lot of atoms will have to be moved and this won't happen overnight. For instance, even if you have a developed a new technology to produce electricity more cheaply, you still have to actually build the power plants that use this technology. Do you think AGI will develop versatile robots that can replace humans and build such infrastructure much faster and more cheaply than we currently can. Great, except that we'll still have to actually build those robots, which means building factories without their help. And even for AGI to develop those robots, it will need data that doesn't exist yet and will have to be collected, which will take time because even if I'm sure that AGI will also help with that it won't eliminate all bottlenecks and the ways it will come up with to help will also require moving atoms around, which takes time. Again, this is true even if you assume away legal and political obstacles, but of course they will exist and be massive. Saying that doesn't make me an "AI skeptic", I'm actually quite bullish about AI. I have no doubt that AGI will eventually happen, though I suspect that most people in the AI industry overestimate how quickly it will happen (but that's a view I hold with relatively low confidence because I don't think anybody knows), and I'm certain that even short of AGI, AI has the potential to profoundly transform human civilization. I even think we should take seriously the possibility that it might end it, so I'm hardly a "AI skeptic" in any meaningful sense, but I'm increasingly annoyed by how sloppy and intellectually lazy many AGI-pilled people are. They constantly make very strong and highly non-obvious claims, such as the claim that the "AI race" has a winner-take-all dynamic, without a clear and rigorous arguments for them and dismiss anyone who asks them to produce such an argument as an unenlightened soul who hasn't seen the light yet. As Ruxandra notes, they clearly think that if you disagree with them it's because unlike them you aren't privy to the secrets of the gods, but no actually that's just because they haven't thought things through. If they had, they wouldn't think that things which are most definitely not obvious, such as the idea that the "AI race" is a winner-take-all kind of things, are obvious. To be clear, I'm not saying that it's not a winner-take-all kind of things, only that it's not obvious that it is and that if you think it's obvious this only demonstrates that you have not thought things through. That's one example but there are many others, such as the claim that restricting access to Chinese open weights models is obviously justified because they will make companies like OpenAI and Anthropic unprofitable, which doesn't actually follow even if you accept that already non-obvious premise. One reason this intellectual laziness and sloppiness annoys me is precisely because I'm not an "AI skeptic", but actually think AI may be the greatest thing that ever happened and I fear that it will take longer for us to reap the benefits it will bring because we don't think about it carefully enough and end up adopting bad policies or unnecessarily increasing opposition to AI as a result.