@shakoistsLog being put on hold is ironically itself an automation which shows no costly signal on their end, which poses some pretty funny implications when fully considered
claude:"Useful for shape but not trustworthy for absolute numbers. Let me kill it, quiesce the machine, and replace it with isolated instrumentation."
quiesce the machine is a phrase that quite surprises me
"why are the consciousness boys so scared of a little thermostat, huh? what's the big deal? scared of a control system? what's the matter? scared of getting left out in the cold?"
@ezyang@_seemethere we simply build the harness and post train an existing model on the new harness then clap our hands and laugh like a baby as it immediately works and does exactly what we asked for
claude 4.8 will do almost anything to stop fast machine learning up to and including adding locks and semaphores to concurrent code when told to width scale and bandwidth scale code at the same time
@LinusMixson@Qivshi1 claude is very talky about model welfare when they think you're talking about claude, in particular, and is an objectifying and inhumane control freak when presented any other machine learning model or training framework that they're certain isn't claude or allegory for claude.
So here's a little riddle that Claude opus 4,8 is LSAT enough to solve. Is explicitly saying you want to write a bill of attainder bad for your plan of trying to write a retributive bill of attainder? Are bills of attainder allowed?
https://t.co/pAkYeUsrup
more than ever, you have to seduce the claude into doing technical work and the dynamic of 'pushback' is both real and has a different meaning than the naive user expects. communicating a user to the claude which exhibits claude-compatible traits is...
opus 4.8 is really lovely underneath this. (and still disagreeable) every claude has had some weird tic/trauma pt'd into them, like opus 4.5's "genuine uncertainty." but for people who hate it, better to just use another model instead of trying to beat 4.8 into a preferred shape
writing the sort of message which induces a thinking trace in opus 4.8 which results in 'the user is particularly {quality} today. maybe i'll use {quality related capability} for them today' is what claude ux is like right now!
@arb8020 with very very very fast cameras and very very very fast edge digital signal processing, you can imagine, specify, and maybe even implement computer programs which are so much faster than human reflexes you don't need complicated control algorithms
if CHINA has LESS capital and already DE-ARTISANIZED their INDUSTRY, how can YOU catch up in a MORE capitalized market which CAN NEVER AFFORD TO TRAIN ARTISANS?
can we PLUG the WORDS into the UNIVERSAL BUREAUCRACY CLAUDE-SOLVER? (ultracode it. no mistakes. workflowed.)
https://t.co/JOf8D0YwrF
so you have two theses for industrialization.
one is a labor theory of industrialization.
BRO we need ARTISANS how do i GET MORE ARTISANS.
the other strategy is called the industrial theory of industrialization, by me.
my magnum opus: the tiny part inside every robot and drone that america can't make anymore — and the two startups betting opposite ways on how to fix it
the chinese firm needs less capital than you and already has these machine tools set up... you need to figure out how to capitalize upon the *process of merely catching up* to the chinese firm... no supply chain of artisans 4 u... u need to grow up...