cc is salty because lesser software has "constraints"

incredible that the only way to predict what a browser will show is literally try every combination

legal and moral grey area but lol, cool hack

knowledge work : boring, probably a commodity
cognitariat : prob a commodity still but has read classics

Pigouvian tax - Wikipedia

en.wikipedia.org/wiki/Pigouvi... TIL : a tax on negative externalities

cynically, most (all?) findings have a best-by date anyway 🙃

Crowbar - Existence Is Punishment YouTube video by Cannibal Nicorpse

youtu.be/mM2-iPAY2p4 crowbar intensifies

"consent" popups exploit a legal loophole. Problem is, lawmakers didn't have the spine to actually ban tracking cookies, so now we get this

always forget I have 2 groups of DNS records: where the website runs and where I registered the domain.

The name servers need to coincide 🥶

brought up the old "special projects" dot com once more

indeed designing with all that yaml is a breeze now, but TLS termination is hard as ever 🥶

I do a similar thing but use k8s deployments as sandboxes. the broker service is an interesting design, will you open source it?

data governance 🤢🫸
data dumpster diving 👈👈🤠

I think the overage is billed with api pricing instead 🥶

actor and spectator, very postmodern

is it a eurovision thing?

whew, that is certainly a take.

it's a good abstraction, but as always the git docs do a terrible job at explaining it, and CC opted for some bizarre defaults (like checking out main and not your current branch for the subtree)

no need to apologize! we're hear to chat and learn new stuff ^^

what I mean to say (claim!) is that public funding can achieve comparable levels of system reliability without centralizing all data flows into a single hub

agreed this need to be an open thing. Not going to rewrite our auth layer otherwise 🥶

all while paying for it a fraction of the median anthropic plan, I'd suspect.

to make this a more pressing case, I shudder thinking of a water reservoir or electric utility that is out for 7 hours in 3 months (like anthropic recently)

and tbh, the interaction of heavy traffic, caching and garbage collection bites _everyone_ in the a++, but rage baiting for clout is so much easier

*cause and fix

nothing posted, I only make terrible noise with it 🥶 I followed a similar arc in fact, but slowed down due to the absurd cost of the modules

idk I'd jump in the sea and forget about coding for a few hours 🌊

hello fellow eurorack person🤙

Turns out if you take the weights and tokenizer and directly materialized them to static globals at compile-time, inference gets very fast!

Very very fast!

it's also hard to gauge the effect of web searches that take place during generation; another factor paywalled models like GPT can't be controlled for.