don draper pitching the Archons on the market research showing that the 18-25 demographic is returning to the Monad at rates not seen for centuries

Timothy Snyder and Tooze are alike in that they both wrote very good books about central and Eastern Europe surrounding WWII and have basically good politics today but have become cranks around the edges due to an obsession with abandoning the US in very particular ways

i mean, sure, but that's partially for the same reason that leftists could not have caused Harris to lose the election: there are not enough leftists to do that.

the thing is that it's not true but the problem is that the dumbest leftists are constantly saying it is. just a shared substrate between morons on both sides.

rhumb lines!

no, it doesn't do that. it, literally, is just an analytics platform which ingests data from elsewhere.

oh so not as KV cache prefill either? are you just doing... live activation patching?

oh wait are you injecting this through the Gemma 4 embedding skip thing?

if i were to put on my "reasonable causal stories" hat i would point out that television isn't real

a combination of antisemitism and the impossibility of it being anything other than naked sectarian malice

this sucks

the Falklands war was an act of imperialism by a European country (Argentina, 97% white) against the indigenous people (the English) of the Falklands and their suzerain, a country with substantial populations from the Global South (the UK, 86% white)

falklands je srpska

spending all my training time and Lipschitz budget on turning x^2 into 1/f before i even produce a single comprehensible pixel

right, yes, i suppose that Word Perverts are training models which aren't conditioned so badly they implode if you try to train them in bf16, whereas us poor diffusion model guys would rather be training in float64 to find a gradient.

this is why LLMs are sandwiched inside an embedding model which tracks word similarity -- it's the only way you can get a gradient at all in what is fundamentally a discrete selection process.

basically, argmax is a 1/0 loss. it's either "got it right" or "didn't get it right." step function. you can only use combinatorial methods there, and none of them are any good. to get a gradient, you have to smear out the words across multiple different related candidates.

actually the other way around. argmax is nondifferentiable, whereas softmax is, so the model is trained on a softmax objective but, during inference, can produce argmax without much of a problem.

the best argument for feminism is to just let an old woman talk at you about her career experiences

"here i made this machine out of discrete log-uniform distributions approximating higher-degree nonlinearities and it works well enough to tell you your paper sucks"

yeah i just generally have problems with mathematical disproofs of things which i have existence proofs of

oh good so we can just quotient them out and go back to things being continuous

machine learning papers assume a lot of things are measure zero for a field which exclusively works in 32-bit floating point numbers, where measure-zero things have an actual width

also, "moses" has an egyptian theophoric name particle ("born of") which never occurs on its own for a name. i would bet money that his name was "el-mses" or something similar, restemmed by people who were illiterate in Egyptian who read it as "mighty moses" rather than "born of El."

many americans demonstrated repeatedly that the thing that would get them to participate in politics was to be told by a television celebrity that they could be rich and it was good to abuse the weak

In the same manifesto, Palantir's owners argued that we need a national military draft, that soft power is over, and that we were too hard on Germany and Japan after World War II. I don't think that company should be allowed to exist anymore.

bibi is literally a holocaust denier and makes common cause with antisemites because antisemitic leaders make new israelis.

so has israel. every presidency since 2000 has involved israel lying to us, browbeating us, attempting diplomatic sabotage, and attempting to get americans killed so that we are forced into a war.

israel was the "cause" of the war in the same sense that the assassination of archduke ferdinand was the cause of world war i, but absent that cause it still would have happened. sort of eggshell plaintiff rules.