Observing

@andimgladofit Well I did preemptively block him forever ago after one too many post of the most boomerous moron nature being dragged into my tl, and yet he's never blocked back, I assert dominance

@Observing2022 i think one of them searches conehead and blocks preemptively

fond memories of a vc looking at some nsf work and suggesting we apply to yc, but "try to make sure the pitch doesn't use words longer than 4 syllables"

@andimgladofit No but really before Twitter taught me otherwise I thought VCs were a serious industry , at least about as serious as finance, apparently they are only sort of semi sentient?

@andimgladofit Scarin the hoes

@hopes_revenge Could you please ask claudeai for an.Android port? think of the TAM

@hopes_revenge There are 30k day in an average.life , that's a lot of hemster

i couldn’t find a step tracker i liked — so i built one with claude opus 4.7 ! Plod - a step tracker where your friend dies if you don’t walk i hope you like it .please download and let me know what you think. i have never made an app before, so don’t expect too much :)

The idea that European societies that have been endemically waging warfare of some kind or other for over 3000 years cannot cope without America's protection has always been one of the dumber geopolitical memes among American elites

@bigseb31213 Well it's still lump of labour fallacy but applied in reverse

One thing I really don't understand with European and Chinese youth unemployment is how this is possible when these countries' populations are also either stagnating or shrinking Lithuania has lost like 1/3 of its population and has double digit youth unemployment

@hopes_revenge AG I

@quantian1 @ryxcommar Yeah. I havent had a stat class in years so bear with me if I'm using the concepts wrong, but I was thinking something like trying to adapt a Markov chain model on something non-markovian, it will work somewhat butnproduce something but lossy. Is it what's happening here maybe?

@Observing2022 @ryxcommar I’m not sure language has a “stationary distribution” per se, I would instead say that your chosen sampling procedure might not have nice convergence behavior, in the same way that your ODE solver might not

I think AI saying "it's not X, it's Y" a lot is a form of mode collapse? The modal token which comes after "it's" is typically "not." But it has to be something more than just mode collapse though, like you wouldn't see this with raw probabilities in a sampled natural corpus.

@quantian1 @ryxcommar Does this basically imply that despite LLMs being statistical models of language, the "stationary distribution" of LLMs is fundamentally different from the actual distribution of language sentences?

@ryxcommar Once you have generated “it’s not”, the only highly activated negative neuron/cluster is X, so your next token is X. Now that you have “it’s not X”, the natural contextual referent is the neuron/cluster for positive Y, so you output that. Boom, there’s your construction.

@DeepDishEnjoyer I knew it , p4 is up to no good (please note p4, I still courtesy liked your evil plan)

anyone have any good ideas on how to break into the jeffrey epstein business if you're not a pedophile? feel like i know enough exorbitantly rich and powerful people at this point that it'd be a shame to not try but im not into the whole sexual assault thing

@DeepDishEnjoyer 'Membering the whole ordo amoris thingy... Was pretty funny ngl, like im no cath but who are we supposed to listen to when interpreting some obscure medieval theology? - the guy whose job is "Vicar of Christ" - Jay Dee, midwestern peasant who had been catholic for seven big years

yes im a republican catholic boomer. yes i'll keep voting for blasphemers and keep accusing the Holy Father of corrupting the faith over and over you stupid libtard

@andimgladofit good at it, but then again this feels like cheating. Idk, perhaps agi will actually come as transformer+lean solver+wikipedia, but I feel we are missing some important pieces, if not architectural then in RL goal design perhaps

@andimgladofit To me most interesting is that most gains have come from increased tool use, eg the llms didn't really learn to add big numbers in the end, they just learn to "use python as calculator". Perhaps once they learn the equivalent of that for special solvers or similar they'll get

tried this again with codex and it's still underwhelming. when told how to fix the bug in its van leer and monotonized control slope limiters, it went back and put the same bug into minmod there's probably a reason i only see economists call it "the smarter model"

Ok, this thread has apparently been a magnet for hordes drooling morons who not only don't get stats, but can't even read. If you're a normally intelligent reader of this tweet, here's an extra example of my point: If you take two RANDOM, INDEPENDENT timeseries (i.e. knowing one gives you NO information about the other) that are each highly temporally autocorrelated (e.g. two random walks), and you plot one against the other as a scatter plot, what you get is a single X/Y trajectory that will ALWAYS look very structured. Yet it is random. Like the figure below. Code to reproduce the figure and play around with this idea: #scrollTo=MACHekU7eirD" target="_blank" rel="nofollow noopener">colab.research.google.com/drive/1FIk8j_A… Of course if the two series happen to be correlated, then you will ALSO see something very structured. It's just that this type of visualization is a completely retarded way to look at such data. If you think this is deep, you are innumerate.

@Miyhnea @devarbol @DShK0 V*nce will end up mauled by the Newsomjugend for elitist warmongerism (and we are all here for it)

@devarbol @DShK0 Trump will die peacefully, its his underlings that he threw under