I don’t like the description “AlphaTensor can not only rediscover human algorithms like Strassen’s algorithm”. (Glancing at the DM post, they also seem to talk like that, but I think the OP shouldn’t repeat stuff like that without redescribing it accurately, as it seems like propagating hype.) To me the important parts of Strassen’s algorithm are
1. the idea of recursively dividing the matrices,
2. the idea of combining some combinatorial rejiggering of the matmuls at each stage of the recursion, and hence the idea of looking for some combinatorial rejiggering,
3. and the combinatorial rejiggering itself.
It’s far less surprising (and alarming and interesting) to find that AI (any kind) has improved some combinatorial rejiggering, as opposed to “discovering human algorithms”. (See also johnswentworth’s comment.)
(Glancing at the DM post, they also seem to talk like that, but I think the OP shouldn’t repeat stuff like that without redescribing it accurately, as it seems like propagating hype.)
I don’t like the description “AlphaTensor can not only rediscover human algorithms like Strassen’s algorithm”. (Glancing at the DM post, they also seem to talk like that, but I think the OP shouldn’t repeat stuff like that without redescribing it accurately, as it seems like propagating hype.) To me the important parts of Strassen’s algorithm are
1. the idea of recursively dividing the matrices,
2. the idea of combining some combinatorial rejiggering of the matmuls at each stage of the recursion, and hence the idea of looking for some combinatorial rejiggering,
3. and the combinatorial rejiggering itself.
It’s far less surprising (and alarming and interesting) to find that AI (any kind) has improved some combinatorial rejiggering, as opposed to “discovering human algorithms”. (See also johnswentworth’s comment.)
Yeah, my bad. I’ve edited the post to clarify.
Cool, thanks.