Reactions from ranked influencers
5 posts
the issue is this would imply only monopolies are accelerationist, as if only one lab is entitled to do model training which is super dangerous (if you have 5 frontier labs constantly subsidizing to remain competitive capex also increases) they imply that only 2 labs should do it like national (? 馃槃) AI assets
Fair but tbh the previous status quo was pretty accelerationist (open models behind by half a year or so, still allowing for interesting research on top of that, while frontier is high margin with high R&D spend for massive scaling) The long tail doesn't have to be everything tho Imo it's slower if open models are the frontier and the profits get disturbed between many more players which makes large moonshots harder by default Again I think things slowing down in that sense is great assuming we manage specific risks that come with open models
@BlackHC @martin_casado imo it's absolutely not different to the case where you have 5 closed frontiers vs 2 closed 3 open frontiers because those open ones are equally hard to deploy so you use them serverless so it's not a question of closed vs open it's a question of monopoly vs democracy
@BlackHC @martin_casado this creates a super one-directional research ecosystem that is more or less an echo chamber which imo is counterproductive for research in a country like US where long tail research contributions from everywhere and monetized for short term monetization it will be even worse
Not necessarily but it helps to not commoditize models overall where only compute providers make money The most accelerationist is probably a monopoly that does invest all its profits into achieving ASI and doesn't care much about anything else Note that I think one should support open models for that reason bc the competition you get from open models is more likely to focus on shorter term economic value