[GoLUG] LLMs can't reason

[kc-golug at chadwicks.me.uk]

10 Jun 2025 21:12:13 Steve Litt <slitt at 444domains.com>:

> I could easily make the argument that human beings can't reason.

Turinh said something like program can exhibit intelligence. I believe the crucial word is exhibit. An Art gallery manager might not be able to create Art. The point is creating or engineering something new is not computable. LLMs can't even automate well as they're unreliably probabilistic. They cause problems in code bases as the project size increases even when simply plagiarising hopefully open source code.

This doesn't mean they don't have uses such as notebook lm making a pdf easy listening but we would be far better off if all that money went to something like getting to deep hydrogen deposits on Earth in my opinion. Superintelligence is a complete lie.