Either you are yourself already a better judge of the thing’s quality, in which case the response can be of no use to you, or you are a poor judge of the thing’s quality, in which case you will be blind to the flaws in the synthesized opinion handed back to you.

Whatever your feelings about AI writing (I make a point of not using AI for writing, myself), it's silly to claim Rao is not the one making the point here. The piece links to the transcript of the chat session that generated it, where after having the LLM ingest the paper, he prompts it:

> ‘I want to make the following critical argument: “Have you met the median manager with a few reports or median faculty with a few grad students? Not your work, not your cognitive debt. Otoh, this sort of framing entirely misses the new kinds of work you are doing, which is often illegible or even invisible. A curious factor affecting AI discourse is that most people researching and commenting about the second-order effects have “individual contributor” mindset. Even a modest amount of supervisory experience demonstrates the difference between thinking vs monitoring/supervising, and between being responsible for getting thinking done versus accountable for it. I use AI a lot and it feels indistinguishable from supervisory work to me, with similar cognitive effects. If you manage too much versus doing your own hands on work, you become… a manager. How good you are depends on your aptitude for management work.’

Venkat argued that the study subjects did poorly not because LLMs were dulling their minds, but because the study subjects were freshman students with no management skills being tested on tasks that required delegation, quality gating and exception handling. The study put people in a situation that created role confusion and concluded that the poor outcomes were due to "cognitive debt" induced by LLM use.

[1]: https://contraptions.venkateshrao.com/p/prompting-is-managin...

> "A cache-aware model router will take that into account by adding stickiness to the initially chosen model and keeps querying it."

I'm persuaded that just about every language is "homoiconic" and that _bicameral syntax_ is the actually-interesting thing about Lisps : https://parentheticallyspeaking.org/articles/bicameral-not-h...