kristjansson
Born on February 17, 2015•4614 Karma
This is the right definition. LLM outputs have undefined truth value. They’re mechanized Frankfurtian Bullshiters. Which can be valuable! If you have the tools or taste to filter the things that happen to be true from the rest of the dross.
However! We need a nicer word for it. Suggesting someone has “AI psychosis” feels a bit too impolitic.
Maybe we reclaim “toked out” from our misspent youths?
e.g. “This piece feels a little toked out. Let’s verify a few of Claude’s claims”
[1] here I don't mean to imply agency, just vigor.
This cycle is hostile in lots of ways, but the trustworthiness and absence of hostility in this dimension is quite nice.
What does one do when a full editor consumes too much bandwidth^H tokens? Use ed, the standard editor!
In particular, it's designed for the teletype era, when (a) the user would have a trace of all the commands they'd sent and output they'd received, since it was literally printed on paper, and (b) output was literally printed on paper, and so had a direct, non-negligible cost.
This is more or less exactly the situation LLMs find themselves in. they can attend to ~all the prior output in their context window, but there's a direct cost to adding new symbols to context.
We've got a tool for exactly that setting, so it would be fun to try it!
obviously $2.5e9ish/yr is substantial in absolute terms ... but that's it? They intermediate half the internet and only capture $7m/day?
Sir, that is price/revenue ratio
(jawdrop)
CloudFlare is honestly still iterating to find a moat other than 'really cheap'.
High P/E means a good moat.
(Too high and everyone is looking for a start-up to eradicate your product segment of course).
Sent me to the shelf, but one has to appreciate the word choice. Evokes the peanut oil spilling everywhere, the reach for geologic terminology captures the lithic aspects of the peanut butter underneath.
The massive scale is all massively parallel: test-time compute for users, test time compute for RL rollouts (and probably increasingly environments for those rollouts), other synthetic data generation, research experiments, …