Conversation

Things I wish I had learned at university: - measuring the efficiency of an algorithm in watts - difference between classical math and computation - relationship between geometry and number theory
6
88
Replying to
In practice it often depends on things like implementing your multiplication with a lookup table. I think the Landauer limit would matter if we did best effort computing and could harness the actual amount of determinism we can get from the substrate. We are far from that.
1
4
Replying to
I recall seeing some literature on a concept called “reversible computing” that seemed to be about this too. I guess that’s the synthesis side of power efficiency analysis.
1
2
Replying to
Reversible computing builds on the realization that entropy results from deleting bits. By building gates that permute bits only, you may be more efficient (but you'll still have to flush out the superfluous bits at some point). Best effort computing is largely unrelated.
1
4
You’re unable to view this Tweet because this account owner limits who can view their Tweets. Learn more
You’re unable to view this Tweet because this account owner limits who can view their Tweets. Learn more
unsure about the performance possibilities but I think it's more focused on antifragility—in one example he shows introducing a virus and then a kind of white blood cell. the other benefit is that you can build really big systems that might be partially distributed
1
You’re unable to view this Tweet because this account owner limits who can view their Tweets. Learn more
Replying to and
Your ideas have put a number of thoughts in my mind. For instance, what is the smallest probabilistic compositional reward optimizer? What priors should it get for compositional self organization? Does it work better with centralized infrastructure that spawns an architecture?...
1
Show replies
To run arbitrarily close to the efficiency limit you have to run arbitrarily slowly. That’s a problem... Second problem is that you become ever more susceptible to thermal noise. In principle ECC fixes this — at ever growing complexity.
2
1