[jhbadger]

The idea that LLMs are useless "bullshit machines" is very 2022. We live in the world today of 2026 where Donald Knuth and Terrance Tao, neither of whom have any patience for hype, use LLMs to help them craft mathematical proofs, I get not liking AI, and getting more satisfaction by doing things oneself. I get frustration on how the AI boom has caused the RAM and graphics cards we want to buy to become unavailable/unaffordable. I get concerns over how such technology is being used by the police and military. But in 2026 the attitude that they are useless "bullshit machines" is as absurd as the articles in the early 2000s that still claimed the Web was just a fad that could be safely ignored.

[shric]

Your entire reply seems to refute that they’re useless yet nowhere did it say they were useless.

In fact, in many occasions they point out they’re indeed useful.

[rootusrootus]

They’re not useless, obviously, but they are still definitely bullshit machines. Both can be true.

[anthk]

Knuth has tools to spot and check if the output is bullshit or not. The rest of the world... just no.