Why do AI labs seem to split up more often than they merge? Intuitively, I’d expect the opposite given the economies of scale of large training runs, and the great pressure to have the best AI.
What splits do you have in mind which are so much more often happening than mergers? We just saw Scale merge into FAIR, and not terribly long before that, Character.ai returned to the mothership, while Tesla AI de facto merged into Xai and before that Adept merged into Amazon and Inflection into Microsoft etc, in addition to the de facto ‘merges’ which occur when an AI lab quietly drops out and concedes the frontier (eg Mistral) or where they pivot to opensource as a spoiler or commoditize your complement play. So I see plenty of merging, consistent with the difficult economics of companies racing to AGI. Maybe you’re just paying too much attention to the fun popcorn-worthy human interest stories like ‘ex-OAer launches startup’.
Why do AI labs seem to split up more often than they merge? Intuitively, I’d expect the opposite given the economies of scale of large training runs, and the great pressure to have the best AI.
What splits do you have in mind which are so much more often happening than mergers? We just saw Scale merge into FAIR, and not terribly long before that, Character.ai returned to the mothership, while Tesla AI de facto merged into Xai and before that Adept merged into Amazon and Inflection into Microsoft etc, in addition to the de facto ‘merges’ which occur when an AI lab quietly drops out and concedes the frontier (eg Mistral) or where they pivot to opensource as a spoiler or commoditize your complement play. So I see plenty of merging, consistent with the difficult economics of companies racing to AGI. Maybe you’re just paying too much attention to the fun popcorn-worthy human interest stories like ‘ex-OAer launches startup’.
A lot of splits happen because some employees think that the company is headed in the wrong direction (lackluster safety would be one example).
The splits are mostly due to disagreements, not economics.