Computational Cosmetologist @dferrer.bsky.social · 4h“Anal Trickster“ oh so I guess we’re just digging up my old tumblr profile huh 070
Computational Cosmetologist @dferrer.bsky.social · 01/10/2026The answer it got from this is also a delight. And this was only 18 months ago! 0130
Computational Cosmetologist @dferrer.bsky.social · 01/10/2026i once got this nonsense out of DS Qwen 2.5 14B. These little models could reason ok but the world knowledge *really* suffered 2332
Computational Cosmetologist @dferrer.bsky.social · 18/09/2026Increasingly I see there is but one important normative Truth: "But no galaxy brain stuff, alright?" shall be the whole of The Law. 1608
Computational Cosmetologist @dferrer.bsky.social · 28/08/2026Breaking CAT RULES never results in any action directed at the cat. She instead bolts to find me or my wife to report. If she can’t find us, she will hide far from the cat until the situation blows over. Pet photo tax: 080
Computational Cosmetologist @dferrer.bsky.social · 14/08/2026Life hack: Going to the shark beach to avoid the crowds … and I’m a little whiny southern boy who can’t tolerate northeastern water even in August. 010
Computational Cosmetologist @dferrer.bsky.social · 07/07/2026>fable writes test to assess reach of mundane code bug involving transmission of data between neighboring cells in a grid >names this test [bug]_t_cell_blast radius >”Fable 5’s safeguards flagged this messages” 0490
Computational Cosmetologist @dferrer.bsky.social · 28/06/2026GANs the undefeated top-tier of making images that completely unlike any natural photo but also unmistakably convey *something* from the real category. These examples of “giraffe” have been burned into my nightmares 191
Computational Cosmetologist @dferrer.bsky.social · 26/06/2026Doing politics the right way gang rise up (to reconsider the choices that brought us here) 010
Computational Cosmetologist @dferrer.bsky.social · 26/06/2026Either: + This quiz has no idea what Klein’s positions are + *I* have completely misunderstood Klein’s positions + we weight the importance of certain positions very differently (“Much like Ezra, you also think alt girls are neat, the most important criterion”) + the matching they’re doing is bad 240
Computational Cosmetologist @dferrer.bsky.social · 24/06/2026So ohio fr fr We will never see the light of heaven. 170
Computational Cosmetologist @dferrer.bsky.social · 04/06/2026Looking back through a slide deck for work and damn I had forgotten what local models were like 18 months ago. Task here was to generate a “poetic sounding sentence under twenty words, with three invented fictional words and one real word spelled incorrectly” This was attempt 2. 72B dense model. 120
Computational Cosmetologist @dferrer.bsky.social · 30/05/2026Plus, those are shallow stairs with hard edges co catch on. Now do: 100
Computational Cosmetologist @dferrer.bsky.social · 28/05/2026This, from the DALL-E paper, is what convinced me to give up on gans. Top is real, next is DALL-E, rest are then SotA gans. Looking back, I find the weird horror charming. It’s like something you’d see in a dream 021
Computational Cosmetologist @dferrer.bsky.social · 27/05/2026“We need IsEven to cover all 32 bit values” 0111
Computational Cosmetologist @dferrer.bsky.social · 21/05/2026Choose your fighter: Left: the single-floor JPM Office in Nashville, TN, America's Hong Kong, economic and cultural juggernaut Right: the 35th largest building in the world, JPM's HQ in a up and coming American city not many people have heard of. Does anyone wants to live here? We'll see. 151
Computational Cosmetologist @dferrer.bsky.social · 07/05/2026Had to look it up because I didn’t remember you saying anything like that, and damn that second “quote” is either “best […] movie […] of the decade”-level sophistry or a critical reading comprehension failure. Either option is pretty discrediting. 060
Computational Cosmetologist @dferrer.bsky.social · 09/04/2026Getting this paper sent to me by a number of anti-AI associates and it really seems like there are some critical flaws for the amount of claimed effort by the authors: - The control group should be given a calculator. Otherwise seems like are comparing tool / without tool. AI seems irrelevant here 120
Computational Cosmetologist @dferrer.bsky.social · 17/03/2026Let no one say LLMs can't generalize 231
Computational Cosmetologist @dferrer.bsky.social · 04/03/2026“The Bitter Lesson” is a fine little essay. Anyone in software/ML/AI should read it. But I see a lot of people run with a maximalist overstatement of it, essentially: “structure and domain knowledge are always a mistake”. This is historically dubious and theoretically ungrounded. A thread: 1375
Computational Cosmetologist @dferrer.bsky.social · 04/03/2026So this is a geodesic under the Fisher Metric (the riemannian metric version of the KL divergence) 120
Computational Cosmetologist @dferrer.bsky.social · 09/02/2026Something odd is happening here (overfitting?). I’m off in a cloud of Spanish-language clusters despite speaking no Spanish and following no Spanish language accounts. Not seeing any obvious followers that are either. But I have a Hispanic last name that’s part of my handle. 000
Computational Cosmetologist @dferrer.bsky.social · 09/01/2026I tried ~30 on openrouter. None of them get it without prompt additions. A lot of models (GLM 4.x especially) do go through the correct meaning in reasoning but decide not to give it. 010
Computational Cosmetologist @dferrer.bsky.social · 28/12/2025Decided to test-drive GLM 4.7 by having it write an ad for this. It absolutely delivered. LLMs are great at right-wing scam ads **for some reason**. [button: DETONATE MY POTENCY] 150
Computational Cosmetologist @dferrer.bsky.social · 19/12/2025I can sympathize with "(small-l) libertarianism, but no galaxy brain shit, alright?" In my insufferable college days, I described myself insistently as a "libertarian social democrat". I stopped saying it more because of the typical reaction than any change in my views. 1100
Computational Cosmetologist @dferrer.bsky.social · 17/12/2025This is a Kerbal Space Program level understanding of the problem. Yeah, at the highest level it could work. It just presupposes a ton of engineering advances and perfection of several technologies that haven’t been demonstrated beyond small scale proof of concept. 110
Computational Cosmetologist @dferrer.bsky.social · 07/11/2025Some of them are so well known that we don't consider them flaws. There is an image that looks *more* like a smiling human face than any real face---despite objectively having little in common with one. We could imagine aliens or AIs without our visual system finding it strange we see this way. 010
Computational Cosmetologist @dferrer.bsky.social · 16/10/2025My completely unserious personal benchmark for new LLMs is how much joy asking basic SWE questions to this pipeline brings me. We've come a long way in the last year. Here is the triumphal conclusion to "The Linterphallus: Symbolic Castration of the Pythonic *Real*" from DeepSeek 3.1 200
Computational Cosmetologist @dferrer.bsky.social · 16/10/2025Apropos of no other discussions of unstable loci of meaning and the size of numbers, here's my first test chat with an old RAG demo I made that uses tools exclusively to find relevant passages from philosophy ebooks I pirated for class in my undergrad. Its takes are much hotter now. 120
Computational Cosmetologist @dferrer.bsky.social · 18/09/2025This may be the most fun I’ve had making diagrams for a presentation 010
Computational Cosmetologist @dferrer.bsky.social · 14/08/2025This was great for early modeling of smaller graphs like social media and molecular structures. But it requires us to *know* the graph in advance. We can combine these two ideas: use a bilinear form to compute the edges from each node's vector embedding, then pass messages on them 110
Computational Cosmetologist @dferrer.bsky.social · 14/08/2025Now, let's say I want to *do something* with a graph. Pre-2017, the way to model interactions on graphs was "message passing on graphs". Essentially, each node is given a vector. We combine vectors along weighted edges, and then mix the results using a standard perceptron. 110
Computational Cosmetologist @dferrer.bsky.social · 14/08/2025But there are other, less regular graphs on language as well. Here we have phrases to their referent phrases across sentences. These kinds of relationships are much harder to compute from a formal procedure. These semantic graphs are derived from the syntactic graph (the sentence diagram) 110
Computational Cosmetologist @dferrer.bsky.social · 14/08/2025The familiar sentence diagram is a graphical representation of the relationship between words. Traditional language AI often tried to algorithmically construct structures like this. At the single sentence level this is tractable. Grammar has a strong, regular structure. 110
Computational Cosmetologist @dferrer.bsky.social · 14/08/2025Language has many parallel graphical structures. The first is the chain-graph of adjacency. Words are made of up of letters arranged in a particular order. For LLMs, the equivalent of this is the sequence of tokens. 120
Computational Cosmetologist @dferrer.bsky.social · 30/07/2025All of the big models released this year can answer this question well without tool calling. We can try to tell ourselves that this is just because the model has memorized every possible description of geographic relations between states, though this isn't a common benchmark or thing people discuss 140
Computational Cosmetologist @dferrer.bsky.social · 28/05/2025A fun hobby project I’ve been working on is seeing how much (with a mixture of RAG and prompting) I can force a model out of the groveling, obsequiousness Instruct-tuning tends to give them. I present “The Lacanalyzer”, a Qwen3 based chatbot with a condescending obsession with critical theory. 161