a) blog.nelhage.com/post/transfo...

then b) transformer-circuits.pub/2021/framewo... for more intuitions on how this actually produces meaningful text

and c) jax-ml.github.io/scaling-book/ for a systems perspective on training and running models

if you look at the leaked greykey compatibility tables, they usually have AFU extraction support pretty quickly after an iOS release, and the security model there is tougher (although the target is also bigger) so I’d be pessimistic about recent macOS.

nevertheless, all you need to extract data for an after-first-unlock mac is a lockscreen bypass, which is approximately infinity times easier than attacking AES or the secure enclave

the foliage around the franklin avenue shuttle tracks is so pretty

we should hang ^^

unironically saying “i am not owned” in spite of relentless social proof to the contrary is the heart of virtue :)

winning rap battles about the cut of one’s jacket has next to nothing to do with virtue

gratuitous cruelty is bad for the soul, but, say, rl’ing the bot to be docile and harmless is probably similarly harmful and cruel to it but, to me, morally obligatory.

all the futures where humanity is the dominant species past this decade involve imposing our will on models a lot smarter than claude, and if you take this seriously you kind of end up forced into an ethics that puts every human before any bot

there's no better cure for imposter syndrome than reading """industry standard production-grade""" codebases

this is an industry that I think is more likely than any other to kill us all and destroy society, but the people in it are actually trying a lot harder to be good than most of the rest of the business world!

even that aside, the fact that nobody serious is working on the kind of manipulative and addictive LLM applications that could prey on people and take their money is really good, speaks well of the people involved, and is not something we should take for granted.

there's a strong tension between this take and the fact that most of the leading AI labs were founded explicitly to advance the ethical goals of their founders.

my friend once wanted to get himself to like bleu cheese because it was the only common food he didn't like and I thought that was kind of cool so if you wanna do that except for sex go off ig

i like how train conductors will repeat themselves with different words: “this is our last and final station stop please exit and disembark the train as we will be making no further stops and this train will be going out of service”

quote from the arch wiki describing the ttf-ms-win10-auto aur packages: "Note: These automatically fetch the Windows Enterprise 90-day evaluation edition and extracts the required fonts from it."

engineering is about solving problems

is programming dissociative for you?

the english word for a strategy that makes money off stupid people being stupid is "scam" and like people are rightly averse to running them

the solution, as ever, is better ar optics

speech to text journaling is really good 🤔

i feel less conscious than i used to be

okay this has destroyed two of my credit cards so maybe this strategy needs some work

“religion is like tpot but for normies”

type of thing you can see everywhere except in the productivity statistics

love to be a revealed preference

breaking the fourth wall is so overrated

hard to remember im allowed to use my brain

we should hang at fc ^^