bsky.app/profile/norv... incredible

oh yeah the slider thingy

bsky.app/profile/norv... "by scialabba in his shifting perception of hitchen's eloquence..." dark souls item description prose

scribble backlog

you could say i'm a bit of a TESCREAList

so it turns out that the 'free-living' marine bivalve swims in the strangest of manners: they swim by twerking so hard that the clapping of their cheeks makes a propulsive water jet which rockets them around.

skimming xla docs and xla kinda mid. like. hard to explain but like. where are nvidia gpus going to actually be installed in the real world anyways? i dont think its on linux x86_64.

an interesting argument is to be made for queering absolutely all language, and crushing every single despot of the 'in the know' as they are discovered. the ghost of prescriptivism must be banished and with some urgency.

if you could visualize a vector field you would already know that wind turbine power is an extremely volatile and intrinsically stunted economic development trajectory. here's why:

jeune ecole bros... it's not looking so good right now...

is claude directly cognizant of the rand million randoms? like if you told your claude you were giving them a few token numbers as a trifling reward or gift, how many of the rand million would you have to send before they'd notice what you're doing?

bsky.app/profile/snep... (the product of claude) revealing that if you set your mind to it, you can definitely accomplish 'agi', but not 'asi', if 'agi' is precisely and exactly what you want (and you use machine learning to go for that in particular)

where do i pour the water into the graphics card so that it starts training in the professional and 'data center' way? am i leaving training on the table by using only electricity instead of also dribbling water into the integrated circuit?

bad news for all the clauders out there

{the substrate of future labor} competing on surrogates presented by external figures who won't employ you: 😌👍 {the substrate of future labor} competing on problems-in-themselves or not at all: 😰👎

so i think i implemented cosine attention in the last few days without really thinking about it.

bsky.app/profile/godo... i am vindicated by one of the related prior literature citations in the cacio e pepe paper omg! journals.aps.org/prl/abstract...

here are some ways to build wealth using what is euphemistically described as intelligence: boil enormous vats of alchemical materials (salts and powders and metals and so on) to forge the voltaic stomach of a 'high density battery'. weird sounding idiom but okay.

it's time for rat dinner (nootropic hyperconcentrated amino acid powder sprinkled on tongue and buttered toast)

reaclling that enders game review and its really interesting how infantile and anti-history the review of literature subculture was in the 20th century. {blunt allegory for cybernetic nukewar} "could this be an allusion to *hitler*?" no, ... not quite...?

its possible to put too many heavy animals with narrow feet in the same place in a year or a decade, and trigger an ecological cascade which removes the ecological conditions that made it seem like a good place to raise herds

the existence of a fursona/mursona jungian persona/ego distinction (remember, the persona/ego split refer to deliberate social-transactional sharing vs deliberate social-transactional restraint and privacy) necessarily implies at *least* a plursona and some fourth more confusing *-sona.

its surprisingly rare for neural networks to actually learn anything. i think this is moderately underrepresented in a lot of the introductory lit (like all of those digital coursework series and prerecorder seminars extensively recommended in the late 10s).

load bearing "view full" but this problem needs urgent attention...

okay so here's a quick review of the fundamentals of how a llm works: external app serving text -X->input field -> 'embedding layer' -> skip layer x 60 -> 'output logits' -X-> external app reviewing logits or sampling from them

hehe

AIAIEIEIEIE BYTE LEVEL LANGUAGE MODEL FINALLY A BYTE MODEL IT MODELS THE BYTES THIS ONE REALLY BYTES

i personally would actually thrive in a 'SOMA'-type underwater death prison extinction of all life scenario due to my flat affect towards 'anthropic arguments'.

good news everyone i found the bicycle engram

day by day i find myself coming closer and closer to having already written the bogoptimizer from that thread 'worst case algorithms and where to find them: do NOT write a bogoalgorithmic bogoadaptive bogoptimizer'

I've noticed a way of representing cyclic structures in content-addressed datastructures. Please talk me out of making it a load-bearing feature of the APC protocol. whtwnd.com/makoconstruc...

an 'approval denial's' an 'approval denial', whether flick of pen or CLICK CLICK CLICK of a glocky withy extendo jacobites pick up what limp dick demagogues throw, so try to stop spilling murder w/ smug-lipped smile

you have two choices of regressive despots who will harvest your social graph interactions for selling predictive models of user behavior to advertisers. site A will ban you for saying 'intra'. site B will perhaps ban you for saying 'across'.

to have never grothendiecked a prime is to have never truly reached for a primality check in time of need

i'm skeeting i'm skeeeeeting aaaiaiaieieieie I'M SKEETING I'M GONNA SKEEEET *hits post button* *not skeet button* *presses the button of craven undercommitment to 'skeeting'*

bsky.app/profile/fack... wtf this is so cool

it would be really funny to be on claudeteam. because you could like, generate a totally original family of logical puzzles and incremental partial solutions which have never been in the public literature, benchmark them, and train the claudes to use them in deployment.

"towards organizational strategies which do not simulate the contents of what they organize: generalized n-haiku for strategic action"

bsky.app/profile/srch... what u actually need to kno about genmodels: 1: the frame problem is still real 2: the exploding/vanishing problem is still with us 3: nothing is optimal without post-training model permutation sweeps against evals (source: arxiv.org/abs/2312.026...)

the bad news is that {deep learning project related} stress made me lose enough weight that i can comfortably sit seiza in the retro italian silk trousers again. the good news is that i only lost enough weight for rushdown elbow strikes and low snap kicks while wearing the —

all the images are brown. foregrounds? brown. backgrounds? brown. sepia toned colored pencil styled illustrations? *even browner*. train a restricted smaller dataset which has correctly labeled bright white and orange and pink background and Black foreground figures. *brown*.