I take back what I said, that "AI" would turn out to be useless in war other than as responsibility laundering to do war crimes. Turns out "AI" can and has killed multiple terrorists. Namely, terrorists who asked training advice to "AI"

https://pivot-to-ai.com/2026/07/13/terrorists-are-using-ai-and-the-chatbot-is-killing-them/

there is thus one way to use "AI" as a war weapon: convince your enemies to consult it

@elilla also image recognition to autonomously guide kamikaze drones into targets which, combined with gps/glosnas/galileo guidance, makes em effectively jam-proof. its kinda scary actually

@elilla thats how ukrainian drones have been able to strike oil refineries in moscow and saint petersburg

@coolbean I do not count image recognition or traditional machine learning as what is now meant by "AI"

@elilla fair enough

frankly i wish it were possible to do image recognition without machine learning anyway but thats just me preferring actually human comprehensible algorithms

@elilla also like... power efficiency. thats good too

@coolbean @elilla rockets have been using optical guidance based on terrain maps before ai. Afaik that's where the algorithms optical mice use come from

@coolbean @elilla You can, tho digital signals processing as a field is a mindfuck that kind of stretches the definition of "human comprehensible"

@scathach @elilla any asic that can handle image recognition is eldritch power beyond understanding to me

@snacks @coolbean @elilla What's really cool are 100% analogue inertial guidance systems, like how are you going to guide a warhead onto a random bridge in bumfuckoblast before GPS existed and before the integrated circuit was invented

@scathach @snacks @elilla i remember messing with the a4e navigation system in dcs and its... so funky. like if you pull too many Gs it starts falling out of calibration and becomes increasingly inaccurate forcing you to rely on tacam on the way back to base and its.... fascinating and i 100% do not want to know how it works thats terrifying

@scathach @coolbean @elilla those will always struggle with drift tho which kinda sucks

@scathach @coolbean @elilla interpretable AI models exist.

it's a commonly cited benefit of ODE nets and KANs, even.

@scathach @coolbean @elilla i wish i had the smarts, time and dedication to get into dsp

@snacks @scathach @elilla i dont. ICs are scary and i dont wanna design or understand em beyond what they look like on an iec 60617 schematic

@scathach @coolbean @elilla GMDH came out of a Ukraine lab in the 90's. Its a somewhat lost technology that grows one layer at a time, fits the current layer, and then prunes all but the top however many winners. Then repeats until stopped. Usually get small networks out of it that don't tend to overfit.

KANs basically turn everything in to huge b-splines, but there's some newer work (FastKAN uses resting bitch faces which apparently were another 90's era thing as RBF networks) about it. One of the funnier things with them is after training them on some problem data, you can try to match the actual basis function parameters back to known math functions. so its been used as a cheeky form of "symbolic regression" where you extract the math formulas out of a problem after the fact.

the black box nature of shitstacking multilayer perceptrons is kind of an open issue from the R&D side, it's just that the VC side doesn't care and benchmark chasing is measuring for that holy accuracy percentage at the complete cost of everything else.

on the other hand i'm a weirdo who cares about things like compute, so i spend most of my time these days in neuromorphics and weird niche shit

@icedquinn @coolbean @scathach @elilla
neuromorphics and shit would be cool if it wasnt so raped to make exotic chips.
i hope these fucking losers get a move on https://fab2.com/
https://hackerfab.org/