Post #3671220

2026-07-08 10:10 UTC

@thomasfricke@23.social out of curiosity, why the non-deterministic approach? Seems like a list of ingredients could be better analyzed with a simple matching program. What an I missing?

Replies (2)

@thomasfricke@23.social 2026-07-08 11:20

@GuillaumeRossolini@infosec.exchange I think this is a glorified optimization. This is normally much faster if you don't need determinism. I have seen this in local weather forecasts, but there might be dragons at the edge cases.

Open ##3671219
@TallSimon@mstdn.ca 2026-07-08 10:24

@GuillaumeRossolini@infosec.exchange @thomasfricke@23.social The Silicon Valley AI grift is what you're missing. A few months ago, I took a 63 hour "AI" training computation done on an unspecified GPU server, and did it in 9 seconds on a single, old CPU core. I got better results, too, with a fully transparent approach. I suspect in this case there is probably a better algorithm to use to train their local, "Small AI". That they are focused on a small, local application is the right general direction, IMO.

Open ##3813454