Interesting . Your take that the meat industry would still be dominant in 2-3 decades only 5-15% of the time makes me curious. This requires that tastier and cheaper meat substitutes are around the corner, or at least available medium-term in the available quantities. This is interesting, but it is not the sentiment I got from people who made future projections of alternative proteins by 2050. (I donât have the exact references in my head, sorry)
For efficiency argument, it was more about âmaking food that uses less land and is cheaperââbut not with the same taste, so not the same comparison, you donât have to take it into account.
But regarding the third section I think we are in agreement: it is worthwhile to support alternative proteins in every case, since having them decades earlier would do a tremendous amount of good.
Just to be clear, my intended belief there was a 5-15% of meat industry taking dominance for 2-3 decades after the introduction of an alternative that is very close or beats meat entirely. Though I do think it is plausible the alternatives come soon, AI technology is advancing rapidly; and I donât believe many peopleâs models are properly factoring in AI technology of our current level being applied to more areas, much less what weâll have in the future after significant advancements. Of course, EA tends to be a lot better at that than other charities.
As an example: Theorem proving is bottlenecked by the annoying but solvable triplet of: data, money to train larger models, and companies focusing on it. Scaling the current methods would hit noticeable limits due to planning/âsearch being hard, but would allow a lot of automation towards proving software correct. AlphaProof itself is then a step above the methods that came before it. This could provide a good amount of value in terms of ensuring important software is correct but is generally ignored or assumed to need massive breakthroughs.
I find it plausible that more systems in the vein of AlphaFold (protein prediction, most centrally relevant to meat) can be extended to other areas of chemistry with a significant amount of time & effort to collect data and design. Thereâs big data collection problems here, we have a lot of data about food but it is more locked away inside companies and not as carefully researched to a low level as proteins. I know the theorem proving better than I do the AlphaFold area, but that gets across my general view of âmany mental models assume too much like we are in 2018 but with single isolated notable advancements like AlphaFold/âAlphaProof/âChatGPT rather than a field with much to explore via permutations of those core ideasâ.
Interesting. It feel like this still requires a lot of effort to make it usable in the context of alternative proteins (plus marketing, developing incentives, fighting the opposition, etc.), but if it works, that could indeed be a good news.
Interesting . Your take that the meat industry would still be dominant in 2-3 decades only 5-15% of the time makes me curious. This requires that tastier and cheaper meat substitutes are around the corner, or at least available medium-term in the available quantities. This is interesting, but it is not the sentiment I got from people who made future projections of alternative proteins by 2050. (I donât have the exact references in my head, sorry)
For efficiency argument, it was more about âmaking food that uses less land and is cheaperââbut not with the same taste, so not the same comparison, you donât have to take it into account.
But regarding the third section I think we are in agreement: it is worthwhile to support alternative proteins in every case, since having them decades earlier would do a tremendous amount of good.
Just to be clear, my intended belief there was a 5-15% of meat industry taking dominance for 2-3 decades after the introduction of an alternative that is very close or beats meat entirely.
Though I do think it is plausible the alternatives come soon, AI technology is advancing rapidly; and I donât believe many peopleâs models are properly factoring in AI technology of our current level being applied to more areas, much less what weâll have in the future after significant advancements.
Of course, EA tends to be a lot better at that than other charities.
As an example: Theorem proving is bottlenecked by the annoying but solvable triplet of: data, money to train larger models, and companies focusing on it. Scaling the current methods would hit noticeable limits due to planning/âsearch being hard, but would allow a lot of automation towards proving software correct. AlphaProof itself is then a step above the methods that came before it. This could provide a good amount of value in terms of ensuring important software is correct but is generally ignored or assumed to need massive breakthroughs.
I find it plausible that more systems in the vein of AlphaFold (protein prediction, most centrally relevant to meat) can be extended to other areas of chemistry with a significant amount of time & effort to collect data and design. Thereâs big data collection problems here, we have a lot of data about food but it is more locked away inside companies and not as carefully researched to a low level as proteins.
I know the theorem proving better than I do the AlphaFold area, but that gets across my general view of âmany mental models assume too much like we are in 2018 but with single isolated notable advancements like AlphaFold/âAlphaProof/âChatGPT rather than a field with much to explore via permutations of those core ideasâ.
Interesting. It feel like this still requires a lot of effort to make it usable in the context of alternative proteins (plus marketing, developing incentives, fighting the opposition, etc.), but if it works, that could indeed be a good news.