s-tier channel and vibes

(i was gonna QP sky, but honestly they seem to be dealing w pileons already, so... gonna let it chill)

3) depending on where you live, temp delta is greater in winter or summer, but because fahrenheit is an interval scale, delta's effect on energy needs (assuming constant COP for heat and cold) is asymmetric, too.

1) link is about cooling
2) thermodynamics cares about temperature gradients (how warm "normal" inside is vs how cold "normal" outside is), so the "rough rule" is... rough.

"we'd run it on the other set of tracks at the switchyard, but someone tied the stock price to those rails."

(i was lost in the materialist sauce. i didn't consider this possibility.)

we'll know LLMs have really made it in research when we have a major journal retraction (sorry, half-hearted corregendum) because an LLM cooked the data to fake results

i think this assumes decreasing returns to scale in tokens? (bad attempt to talk through math here, tl;dr if only new entrants, maybe, but why no substitution from RA to tokens among better resourced?)

maybe i overestimate how muchal learning happened when folks had to copy-paste code from stack, statalist, docs, whatever.

perhaps the right mental model is that the author/researcher is like a PI and claude code/whatever llm is like an RA set to a task. but then principal-agent problems arise, eh?

i still teach a boring, out-of-date class on how to actually write code to manipulate/transform data and perform analysis and most of the learning isn't the memorization of the function names, but rather the building of intuitions about how data works, how transformations work, and what steps mean

yeah, this is my understanding of things - if you can afford the big token subscriptions, it scales like mad, if not, you get a lot of interrupted development

this typology is going to live in my head for a while now

(indexed, yadda yadda yadda, it didnt fit the quote)

What is a man?! A miserable little pile of random variables!