Architeuthis

@ Architeuthis @awful.systems

Posts

6
Comments

238
Joined

3 yr. ago

It's not always easy to distinguish between existentialism and a bad mood.

Architeuthis @awful.systems 16h ago

2d ago

Apparently Anthropic may be about to be on the receiving end of some major banana republic shit from the Trump admin -- Update: Anthropic labeled supply chain risk by DoD.

As far as I can tell it's only on anthropic's word that that's the main issue, DoD just talks about unfettered access for all lawful purposes, which is basically a bend-the-knee-or-else framing, and pivoting away from that to bargaining on particulars will make them look weak, so I guess that's that for now.

Αnthropic being against mass surveillance and autonomous weaponry while in bed with Palantir is kind of if IBM took a stand against antisemitism while spearheading the computerization of the third reich prison system.

Kudos to Dario for stepping off the hype train for one millisecond to admit that using an LLM to control an automated weapons platform is currently kind of out of scope for this technology, I bet that took a toll on his psyche.

And also for pointing out that something can be legal only because the law hasn't yet caught up with the technology.

2d ago

Stubsack: weekly thread for sneers not worth an entire post, week ending 1st March 2026

It's entirely possible he does get that it's a nothing burger but is just being his usual disingenuous self to pull people in.

2d ago

Stubsack: weekly thread for sneers not worth an entire post, week ending 1st March 2026

I mean the whole entire premise (not unique to this post, scoot's gotten a lot of mileage out of this) is shoehorning LLMs into the predictive coding framework mostly on the grounds that they both use prediction terminology and deal with work units that they call neurons, with the added bonus that PC posits Bayesian inference is involved so it's obviously extra valid.

Queue a few thousand words of scoot wearing his science popularizer hat and just declaring the most vacuous shit imaginable with a straight face and a friendly teacher's casual authority.

2d ago

Stubsack: weekly thread for sneers not worth an entire post, week ending 1st March 2026

He isn’t even trying with the yellow and orange boxes. What the fuck do “high-D toroidal attractor manifolds” and “6D helical manifolds” have to do with anything? Why are they there? And he really thinks he can get away with nobody closely reading his charts, with the “(???, nothing)” business. Maybe I should throw in that box in my publications and see how that goes.

It's from another horseshit analogy that roughly boils down to both neural net inference (specifically when generating end-of-line tokens) and aspects of specific biological components of human perception being somewhat geometrically modellable. I didn't include the entire context or a link to the substack in the OP because I didn't care to, but here is the analogy in full:

spoiler

The answer was: the AI represents various features of the line breaking process as one-dimensional helical manifolds in a six-dimensional space, then rotates the manifolds in some way that corresponds to multiplying or comparing the numbers that they’re representing. You don’t need to understand what this means, so I’ve relegated my half-hearted attempt to explain it to a footnote1. From our point of view, what’s important is that this doesn’t look like “LOL, it just sees that the last token was ree and there’s a 12.27% of a line break token following ree.” Next-token prediction created this system, but the system itself can involve arbitrary choices about how to represent and manipulate data.
Human neuron interpretability is even harder than AI neuron interpretability, but probably your thoughts involve something at least as weird as helical manifolds in 6D spaces.I searched the literature for the closest human equivalent to Claude’s weird helical manifolds, and was able to find one team talking about how the entorhinal cells in the hippocampus, which help you track locations in 2D space, use “high-dimensional toroidal attractor manifolds”. You never think about these, and if Claude is conscious, it doesn’t think about its helices either2. These are just the sorts of strange hacks that next-token/next-sense-datum prediction algorithms discover to encode complicated concepts onto physical computational substrate.

re: the bolded part, I like how explicitly cherry-picking neuroscience passes for peak rationalism.

2d ago

Apparently Anthropic may be about to be on the receiving end of some major banana republic shit from the Trump admin -- Update: Anthropic labeled supply chain risk by DoD.

I live in the Balkans, I have br-word privilege.

2d ago

Stubsack: weekly thread for sneers not worth an entire post, week ending 1st March 2026

I like how even by ACX standards scoot's posts on AI are pure brain damage

One level lower down, your brain was shaped by next-sense-datum prediction - partly you learned how to do addition because only the mechanism of addition correctly predicted the next word out of your teacher’s mouth when she said “three plus three is . . . “ (it’s more complicated than this, sorry, but this oversimplification is basically true). But you don’t feel like you’re predicting anything when you’re doing a math problem. You’re just doing good, normal mathematical steps, like reciting “P.E.M.D.A.S.” to yourself and carrying the one.

The most compelling analogy: this is like expecting humans to be “just survival-and-reproduction machines” because survival and reproduction were the optimization criteria in our evolutionary history. [...] This simple analogy is slightly off, because it’s confusing two optimization levels: the outer optimization level (in humans, evolution optimizing for reproduction; in AIs, companies optimizing for profit) with the inner optimization level (in humans, next-sense-datum prediction; in AIs, next-token prediction). But the stochastic parrot people probably haven’t gotten to the point where they learn that humans are next sense-datum predictors, so the evolution/reproduction one above might make a better didactic tool.

He also threatens an Anti-Stochastic-Parrot FAQ.

Here's hoping if this happens Bender et al enthusiastically point out this is coming from a guy whose long term master plan is to fight evil AI with eugenics. Or who uses the threat of evil AI to make eugenics great again if they are feeling less charitable.

3d ago

Apparently Anthropic may be about to be on the receiving end of some major banana republic shit from the Trump admin -- Update: Anthropic labeled supply chain risk by DoD.

Architeuthis @awful.systems

Architeuthis @awful.systems 3d ago

Also being in a strategic partnership with fucking Palantir does tend to make one's stand against mass surveillance seem less than genuine.

TechTakes @awful.systems

3d ago

Apparently Anthropic may be about to be on the receiving end of some major banana republic shit from the Trump admin -- Update: Anthropic labeled supply chain risk by DoD.

archive.is /20260226063523/https://www.axios.com/2026/02/25/anthropic-pentagon-blacklist-claude

4d ago

Stubsack: weekly thread for sneers not worth an entire post, week ending 1st March 2026

Architeuthis @awful.systems 4d ago

I mean, sure, but it's still the CEO of XBOX on her second day on the job throwing her hat in the legendarily sus declining birthrates discourse in service of AI solutionism, it's not nothing.

5d ago

Stubsack: weekly thread for sneers not worth an entire post, week ending 1st March 2026

MicroSlop's new xbox CEO has a background in AI and is worried about birthrates.

Architeuthis @awful.systems 5d ago

Can't wait for her lesswrong handle to leak.

6d ago

Stubsack: weekly thread for sneers not worth an entire post, week ending 22nd February 2026

Architeuthis @awful.systems 6d ago

Using talking points meant for c-suites to a general audience and outing yourself as a complete psychopath, the San Fran CEO Story.

1w ago

Stubsack: weekly thread for sneers not worth an entire post, week ending 22nd February 2026

I the post he keeps referring to Ollama as an LLM (it's a desktop app that runs a local server that lets you download and interface with a local LLM via CLI or http API) so it's possible he's just that far behind in his technical understanding of LLMs that he's fallen to taking the wrong people's word for it.

The post certainly reads like he doesn't even know which local LLM he's using, let alone what it takes to make one.

1w ago

Stubsack: weekly thread for sneers not worth an entire post, week ending 22nd February 2026

That he went from that all the way to it's mostly ok when sam altman steals all your data, misrepresents it and then steals all your traffic is... bad.

At any rate it's definitely good to know that that war crime forensics data project isn't quite the unintentional shambles corey makes it out to be.

1w ago

Stubsack: weekly thread for sneers not worth an entire post, week ending 22nd February 2026

That was a good read.

Corey doc wrote:

It's not "unethical" to scrape the web in order to create and analyze data-sets. That's just "a search engine"

Equivocating what LLMs do and what goes into LLM web scraping with "a search engine" is messed up. His article that he links about scraping is mostly about how badly copyright works and how analysing trade-secret-walled data can be beneficial both to consumers and science but occasionally bad for citizen privacy, which you'll recognize as mostly irrelevant to the concerns people tend to have against LLM training data providers ddosing the fuck out of everything, and all the rest of the stuff tante does a good job of explaining.

Corey also provides this anecdote:

As a group of human-rights defending forensic statisticians, HRDAG has always relied on cutting edge mathematics in its analysis. With its Colombia project, HRDAG used a large language model to assign probabilities for responsibility for each killing documented in the databases it analyzed.
That is, HRDAG was able to rigorously and legibly say, “This killing has an X% probability of having been carried out by a right-wing militia, a Y% probability of having been carried out by the FARC, and a Z% probability of being unrelated to the civil war.”
The use of large language models — produced from vast corpuses of scraped data — to produce accurate, thorough and comprehensible accounts of the hidden crimes that accompany war and conflict is still in its infancy. But already, these techniques are changing the way we hold criminals to account and bring justice to their victims.
Scraping to make large language models is good, actually.

what the actual shit

edit: I mean, he tried transformer powered voice-to-text and liked it, and now he's all in on the LLMs are a rigorous and accurate tool actually bandwagon?

Also the web scraping article is from 2023 but CD linked it in the recent pluralistic post so I assume his views haven't changed.

1w ago

Stubsack: weekly thread for sneers not worth an entire post, week ending 22nd February 2026

Also code helper tools don't even work like that, there's an absurd amount of MCP and RAG based hand holding for the chatbot to even get a grip on what it's supposed to be doing at any given time.

Prompting an LLM with your entire code base isn't really a thing, even though the hype makes it feel like it would be.

2w ago

Stubsack: weekly thread for sneers not worth an entire post, week ending 22nd February 2026

Architeuthis @awful.systems 2w ago

Eh, one might say that going by the broad strokes version while letting the expert do their thing is basically what management is all about, especially if they ignore the part where he wants his version to be light and entertaining.

This isn't about managing subordinates though, this is about devising ways to be complacent about not double checking what the LLM generates in your name.

2w ago

Stubsack: weekly thread for sneers not worth an entire post, week ending 22nd February 2026