[ahartmetz]

Chomsky makes the same point in his book "Manufacturing Consent": Economic papers report more truthfully than most because the truth is what their readers need to make decisions in the real world.

I find that pretty convincing and I do check out such papers occasionally for that reason.

[eru]

It's somewhat funny that the rest of Chomsky's ideology doesn't seem to really take that insight serious.

[seanmcdirmid]

Just look at how wrong Chomsky was about his core subject with the rise of LLMs. He basically invented a whole field that turned out to be a very wrong direction.

[eru]

Yes, though I wouldn't hold it (too much) against him when he was young and the field was even younger. In the 1950s and 1960s and even later, it wasn't clear that LLMs would eventually win at all yet alone win so hard. And Chomsky's early work is genuinely useful. It's just that it's useful formal languages, like programming languages, and has only limited applicability to natural human languages.

[ahartmetz]

I'm not aware of any obvious contradictions tbh. What do you mean?