@meeper @a1ba i've seen the papers where they showed no transformer network can perform simple logic tasks reliably.

whatever productivity you got from using an LLM is because what you were doing is just retracing a solved problem (probably stolen from GPL code at that.)