Post #1637594

2026-04-23 12:41 UTC

"Yes but my artisanal model is trained on..." Bitch, IBM Watson was trained on the most specific and largest bodies of work possible, including confidential health records. And it would recommend sulfa drugs to people who had severe allergies to sulfa. It had more power than your entire rack of silicon that is bad at math - over 16TB of RAM and 2880+ cores. And it's the exact fucking same 'AI' concept. Tokenize, analyze, find statistically related phrase. Same. Exact. Thing.

Replies (3)

@rootwyrm@weird.autos 2026-04-23 12:45

So, no! Your "RAG" doesn't do shit. Your "training" doesn't do shit. The entire fucking CONCEPT is so fundamentally incorrect that anyone serious abandoned it 10 years ago. Because we had a mountain of proof that it CAN'T work. NOTHING can make it work. It is and will never be anything more than the world's most inefficient random string generator. That is all it is *capable* of being. The entire *foundation* of LLMs is objectively incorrect and non-functional.

Open ##1637593
@realn2s@infosec.exchange 2026-04-24 06:17

@rootwyrm I would argue that Watson tried (and failed) to incorporate expert systems. The current "AI" don't and are even worse Just masked by using more data to pretend that it is working. I'm getting more and more worried how people who i thought highly off totally fall into this trap 😱

Open ##1713040
@psistarpsiii@tacobelllabs.net 2026-05-01 14:55

@rootwyrm interestingly, a human doctor did this exact thing to me recently - and she was right! turns out sulfa “allergies” in infancy don’t always mean it carries though in adulthood, babies are just sensitive little bastards. Your point still stands and I agree with you.

Open ##1881686