PLTR used to be a meme stock, then got pumped into a "national destiny stock" over the past two years, with all kinds of money rushing in. I was never bullish, and sure enough it just got wrecked again. But I bet these people won't learn — next hot thing comes along, they'll pile in again, then get wrecked again. Chasing rallies and panic-selling is coded in the DNA. Losing money doesn't cure it.

US software stocks getting slaughtered, people shouting that MSFT is about to be done. Same thing every time — on the way up it's "AI changes everything," on the way down it's "the moat is gone." Same people, same ticker, narrative flips 180 degrees in just two big red candles. When there's blood on the street, most people are busy crafting stories to explain why this time is different, why it'll keep dropping.

Model output is unstable, so you add rules. New rules conflict with old rules, so you add more rules to specify priority. The prompt bloats from 200 tokens to 2000 tokens, and changing a single word could trigger a butterfly effect. Eventually nobody dares touch it, because nobody knows why each rule is there.

A 2000-token prompt is untouchable legacy code. And the root of this cycle isn't that the prompt is poorly written — it's that the direction is wrong: you're chasing determinism at the prompt layer, but prompts can't give you determinism.

Blog is live. An AI engineer whose SBTI score came back DEAD finally has his own turf.