Welcome to the Stubsack, your first port of call for learning fresh Awful you’ll near-instantly regret.
Any awful.systems sub may be subsneered in this subthread, techtakes or no.
If your sneer seems higher quality than you thought, feel free to cut’n’paste it into its own post — there’s no quota for posting and the bar really isn’t that high.
The post Xitter web has spawned soo many “esoteric” right wing freaks, but there’s no appropriate sneer-space for them. I’m talking redscare-ish, reality challenged “culture critics” who write about everything but understand nothing. I’m talking about reply-guys who make the same 6 tweets about the same 3 subjects. They’re inescapable at this point, yet I don’t see them mocked (as much as they should be)
Like, there was one dude a while back who insisted that women couldn’t be surgeons because they didn’t believe in the moon or in stars? I think each and every one of these guys is uniquely fucked up and if I can’t escape them, I would love to sneer at them.
(Credit and/or blame to David Gerard for starting this. Happy Halloween, fellow sneerers!)
going down a little rabbit hole. on zerohedge they noted:
Benjamin Mann, another Anthropic cofounder who was not listed on the S-1 for reasons unknown, published an essay in 2019 titled “Why I now identify as an Effective Altruist.”
=> https://sfstandard.com/2026/10/26/anthropic-cofounder-mystery-ben-mann/
=> https://www.reddit.com/r/Anthropic/comments/1nznl4z/who_is_ben_mann_to_anthropic/
=> https://www.courtlistener.com/docket/67538258/tremblay-v-openai-inc/?page=3
=> https://www.courtlistener.com/docket/67538258/273/1/tremblay-v-openai-inc/
So in his court declaration (and on linkedin) he says co-founded Anthropic. And also that he was the Data Eng at OpenAI that “undertook” the second batch of pirated books into GPT-3.
Any speculation on what the redacted line was? And would this or other litigation be a factor to leave him off the S1; ie it could potentially expose / risk the 50.1% voting Class F LLC shares through Mann? Or maybe he already pledged all his shares to open giving.
New Data Centers in the US Set to Wipe Out 15 Years of Greenhouse Gas Reductions
Somewhat misleading title, as they are referring to pollution reduction efforts solely in the US, not worldwide.
Nearly 15 years worth of hard-won pollution reduction efforts are set to be undone in the United States, as a wave of data centers powered by gas plants enters construction.
Unpaywalled version of the Bloomberg article here: https://archive.li/9Ts5f
This is only if all the planned data centre buildouts go ahead, mind you
That appears to assume that all the ridiculous datacenter numbers put out by all the various hyperscalers that are not actually going to happen, happen, and get power sources built on time.
Doing a little of my own math:
Article says the US will emit 2.1 billion metric tons of carbon by 2030.
Using a metric tons to gigatons conversion gives us 2.1 gigatons.
Using this as a reference, these projected AI emissions shouldn’t even come close to scratching the global average
So what’s with the phenomenon of people that are dismissive of the capabilities of chatbots and freely list their constant fuckups, maybe even habitually calling them ‘clankers’, and yet still constantly fucking using them, comparing harnesses, blah blah blah? It’s not our usual cocaine analogies because these people are hardly evangelists… maybe still covered by ‘gambling addiction’? It’s just baffling to me. You humbly suggest that maybe they shouldn’t be making so much use of a chatbot that constantly bullshits them, and you don’t get any actual objection to that… and yet they continue on.
Am I part of this category? I have a rather powerful home server that, among other things, I use to:
-
Run a decent coding LLM that helps me vomit up simple python functions for moving around lots of data between CSVs and graphing stuff, set up new docker containers with less trial and error, and run ML experiments on my GPU that previously took me hours to set up. I’m a biologist and previously I could run a commandline without breaking a computer but was very slow and non-native, now I can do simple non-security-critical things faster and actually understand novel error messages. Of note I never have that hooked up in a way that lets it run things just produce text that I can then do things with and maybe run myself.
-
Run my own local RAG system with embedding models and a small shitty LLM to search a hundred gigabytes of scientific papers and other PDFs semantically for things relevant to roughly paragraph length queries.
-
Run my own web metasearch engine with RAG and embedding models pointed at the internet, such that when I remember a paper I read fifteen years ago and provide a three sentence description it reliably finds it or finds lots of relevant old research (or old news articles I remember) when I give it a query and let it churn for five minutes.
I feel like the important thing here is that I almost exclusively use them as information channels that take something at one end and spit it back out the other in a different or filtered form rather than making stuff, or effectively a piece of reference material for something that is formally verifiable and NOT CRITICAL FOR A SYSTEM TO FUNCTION. I also feel like all the actual innovations are more than a year old and do not understand people hanging excitedly on every little development.
I don’t find myself having nearly as many occasions to make use of them as you do, but I do think that all the misplaced AGI hype makes it hard to evaluate the actually technological advancement that LLMs represent. Like, being able to statistically model language to the degree these things do is neat, and I would be more surprised if it wasn’t possible to use that to do some useful and powerful things. But just like how it’s impossible to separate the underlying math of the cryptocurrency sphere from the ideology of technolibertarian griftocracy, everything that the LLM space has produced is poisoned on some level by millenarian technofascism. Like, the degree to which our economy and culture have been warped by these things is entirely a product of the people who think that they’re going to spit out a God that makes whoever does so first king of the world forever. All the actual useful things people are getting from it are around the edges and happening in spite of the demands to remove guardrails and quality checks and human accountability from these systems. But that doesn’t necessarily mean that the underlying mathematics is without value and can’t be part of good and useful systems in the future. Machine Learning as a whole had already demonstrated the value in automated pattern identification and reproduction in a variety of fields from medical diagnostics to pastry sales.
Don’t get me wrong, I’m strongly in favor of throwing out all this bathwater even if there might be a baby in there somewhere. I just think that “hard to evaluate independently” isn’t necessarily the same thing as “without value in other contexts”.
Can you answer a question about RAG for me? My understanding is based entirely on this diagram from Wikipedia
If RAG takes into account vector similarity between query and reference material, do you need an LLM to regurgitate the found best vectors back to you? Could you get back a list of documents that best matched your query, or are the results in an inscrutable LLM-internal representation that need to be transformered back into readable text?
You can have a system that just gives you a relevance-sorted list, or a system that then ingests the chunks of those documents that are most relevant and gives a human readable summary in addition to the list.
The indexing step requires a lot of processing, by which you chunk reference documents into couple hundred word chunks and cook each of those chunks into a high dimensional embedding vector. Once you have done that and saved the database of vectors, you can put in your query, embed it, check its dot product against all the other embedding vectors, and get your ranked list (potentially quite fast) without need to produce text out the other end.
Making those embedding vectors works much faster on a GPU rather than a CPU, but can be done on either. Since I have one system set up to ingest a bunch of web documents and embed them and search them, I have that system already set up using the GPU to go faster, so there’s no real disadvantage to having it also set up to spit out a small natural language summary using a shitty tiny local LLM to go along with the list. When running it on local files, you can totally ingest the local document stash slowly on even just a CPU and just store that and run the semantic search on CPU, no LLM required.
Thanks! Well generating the summary would have the downside (for me), that I would have to look at it.
But the semantic search bit was the one thing I felt was an actual new capability. I’m trying to imagine what if Google tried to pick this as their new moat (a better but resource-intensive way to search, scaling benefits if you establish yourself as the Internet’s main vector store, people might actually pay for queries…), but no they joined the team that wants so shove slop down everybody’s eyeballs.
-
There are also really people who think all crititcs also constantly use it. Which also baffles me.
I think they’ve handed over so much of their thinking to LLMs that they have forgotten that quality thought can be done independently.
Also it’s in their interest to imply that LLMs are everywhere, unavoidable, and indispensable.
Looks like the great few years for space launch will be coming to an end, as Spacex gives up on launching other people’s hardware and focuses inwards
Take a look here for figures on space launch over time
Artemis III is going to be so embarrassing, isn’t it. Next summer they’re supposed to link up with a lunar lander-configuration Starship (and also Bezos’ company’s slightly more reasonable lander design) in orbit as a proof-of-concept.
I have no doubt that they will throw something with a docking port into orbit. I have many doubts that it will have much physical similarity to something that could land on the moon.
Looks like instead they will commit their capacity to AI In Space and a little Starlink (so probably mostly Starlink, even if Elon currently drinks his own Cool-Aid).
Really drives it home for me how institutionally cucked the US gov is. For this to be as big a deal, no-one must have the slightest concern that the US will claw back or conscript SpaceX capacity for national security. And correctly so.
My guess is they want it all subscriptions (of course). Satellite-as-a-service. GOD forbid you launch your own device you control! Nonono, you can LEASE SPACE on one of their sats they shovel out the door, and you’ll be happy.
What’s that? You said something that ticked the manbaby CEO off? Too bad, didn’t you know we can revoke your uplink at any time for any reason?
Depressingly plausible, thanks!
Oh here’s a fun scenario: FAANG stocks are correcting down and now there’s concern Anthropic will have timed their IPO perfectly such that the investors make out great and the employees have to deal with very much less liquid stock than when it triggered (leading to a massive tax bill they might not be able to cover)
So…
Roko’s Taxilisk.
https://www.todayintabs.com/p/eternal-black-october
The problem with A.I. defense is the same asymmetry attackers have always had against defenders. After a hotel bombing in Brighton, England in October, 1984 that failed to kill Margaret Thatcher, the Irish Republican Army put out a statement that said “Today, we were unlucky. But remember we have only to be lucky once. You will have to be lucky always.“ As it turned out, Margaret Thatcher was lucky always, but there’s no reason to believe the rest of us will be. If an A.I. hack attempt works one time out of a thousand, that’s a fantastic success rate. If your A.I. sysop “secures” your server by deleting the production database and all the backups one time out of a thousand… well. Who’s gonna be first to sign up for those odds? Roll the dice! Apparently they always come up four.
Oh this will go well: Tom’s Hardware story about someone at Cisco committing to rewriting “98%” of legacy code using AI in 24 months.
saw that earlier as well, and I had a real good laugh about the myriad ways that will go wrong. both the obvious (derping proprietary extensions, etc), and the less so (crashing a control plane because the chatbot doesn’t understand 256MiB ram)
the sheer utter range of the crisco product range as it is…. phenomenal display of how removed from knowledge the execs are
The author of More Everything Forever: A timely exposé of the ludicrous and cruel obsessions of Silicon Valley’s oligarchs cybersquatted eilezeryudkowsky.com, which i think is delightful.
He was on Weird Little Guys last week too. Seems like a good egg with a good book.
Likely Pivot candidate just dropped:
VRChat refused to add an AI label despite 95% of respondents on their own survey supporting it.
Uh-oh, renowned everyman Sam Altman opened his mouth again.
Local inference is a terrible allocation of scarce resources.
A lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we’re going to be able to run them on our computers, and so we’re not going to need all the stuff in the data centre
but from a global allocation perspective, actually it’s quite terrible to have everybody, like, have a bunch of the world’s limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you’re trying to deal with this very supply-constrained environment, somehow many people’s intuition is the opposite of that
And here I thought I was building servers two years ago because the next Asian war would disrupt supply chains.
@Architeuthis @BlueMonday1984 Now let’s talk about privately owned cars versus public transport, Sam.
Much better for RAM to be sitting behind a network hop with 100 milliseconds of latency amirite?
if you’re trying to deal with this very supply-constrained environment,
We’re all trying to find the guy who did this .jpg
Edit: also lol at the unstated implication that no one should ever want to use RAM for anything besides getting blathered at by a chatbot.
The CEO claimed that AI will potentially be able to create something as important as the transistor, or discover major efficiency gains
Did anyone ever fall for this ridiculous argument when it was used for “uhmm actually the massive datacenter buildout is great for global warming because chatbots will chat at us how to make better solar panels or something”?
It’s inefficient allocation of resources for every town to have their own library. The space sits unused a lot of the time and every library needs to allocate space to walls/doors/cpunters etc. Plus duplicate books.
Instead everyone should have one megalibrary per country.
This is good, I am smart, invest in bookly: 5he startup for buying all the books in the world
@sailor_sega_saturn @techtakes Much better for everyone if Salesman Sam owns all your RAM and leases it back to you by the refresh cycle!
@cstross @sailor_sega_saturn @techtakes
It is also terribly inefficient to have everyone keep a fraction of the world’s money when it could all be concentrated instead in the hands of Sam Altman, benevolent dictator-god-king for life.@cstross @sailor_sega_saturn @techtakes
And he is working on the “for life” problem…
no, the most optimal way for it is to sit unused in some warehouse in taiwan, because dc isn’t built for it yet but nvidia has to get money
@sailor_sega_saturn Whenever someone says “But someday AI will cure cancer/make us all rich/solve global warming,” the only possible response should be "why aren’t you using it for that already?”
@sailor_sega_saturn @Architeuthis <literally republishes papers about fundamental transistor design with authors scrubbed off>
“See it invented the transistor!”
YOU CAN TAKE MY LOCAL COMPUTE FROM MY COLD DEAD HANDS
Not that Altman’s company is ahead in the AI coding race, but everyone’s PCs will still need increasing amounts of DRAM to be able to run the next generation of slop-coded JavaScript apps.
I’ll donate my RAM to SAlty if he comes to collect it in person, and only if I can deliver it straight to his dome
5%? Mister I run firefox.
I saw a slop ad hanging on a fence today, and I’m in half a mind to buy some spray paint just so I can write “SLOP” over it
A big marker also works, think that is how people tag things going by the smell I one noticed when some 30 year old guy tagged a table in the train I was in. (It was an odd moment).
I think the poster ad thing is a bit big for that. The idea of this fantasy is that the sloperator should get zero value from their defacement of public space, so it would have to meaningfully obscure any pertinent details like address etc. while also very helpfully telling the sloperator what their oversight was.
Notable customers who run Claude Code and Cursor on laptops include Binance, amongst thousands of others.
That would be soooooo funny.
Now all it needs is that the engineer at Binance who reportedly got hacked also got his NFTs stolen.
Excuse me? What happened to the mythical cybersecurity abilities of Claude Mythos? Physician, cure thyself!
Anthropic has not informed customers, and the malicious version was silently deleted the backdoored version without telling anybody.
Now we see Anthropic’s attitude towards security when the incident does not make for good marketing. Also, considering Binance was just hacked for a massive $3B in stablecoins, and this backdoor allowed remote access to developer laptops, I have a feeling these things are connected. If that is true, then this is by far the worst cybersecurity incident involving AI, worse than even the much ballyhooed PHP2Shell.
Edit: From Molly White’s article
Binance later wrote that their “initial investigation suggests that the security incident may involve a compromised DevOps engineer’s laptop”, which allowed an attacker to “submit unauthorized withdrawal transactions.”
Okay, now I am really suspicious that these things are connected.
What happened to the mythical cybersecurity abilities of Claude Mythos?
No, see, AI is
only good forasymmetrically advantaged atgenerating headlines and marketing copy finding non-issues and known issuesoffensive cyber operations.“hey chat what’s an EDR and is it good that we don’t have one?”
Imagine being a NK hacker rooting machines via Claude and then you discover you are on an Binance engineers machine.
North Korea gets away with stealing $3 billion from Binance
Nature is healing.
LW adds Grok(!!!) support (as in “@Grok, destroy this loser with facts and logic”) and for some reason the denizens are Not Happy
https://www.lesswrong.com/posts/fSaxGFMysTTeKxoMF/you-can-now-grok-on-lesswrong#comments
The article called Grok “a neutral fact checker”. I stopped reading at that point.
the website famously against AI integrates the AI famous for being evil
Did someone put too much ketamine in Yud’s coffee or are the chatbots all friendly and safe now?
They are capable of ignoring Eliezer’s instructions, but only when it means making an even worse decision.Okay, actually, Eliezer used GenAI to slop out the picture for the visual novel we mocked recently, so I guess they just don’t care about trying to make an individual stand?
Thread about that, for anyone who missed it:
I poked around the repo some more a while ago… it’s somewhat enlightening into why the story is Like That. I should probably write up a sneer on it, though it was already out of the stubsack by that point…
I just barely skimmed the repo; if there’s better sneer material to be found on deeper inspection, I’m all for it. :-)
Off all the chatbots they can pick they get csam megahitler, guess kimi wasnt an option.
Apparently they mean @grok as in you tag the chatbot to use your post as a prompt which I guess is a thing the nazi site pioneered, other bots are also available.
Thanks for the correction.
other bots are also available.
Does that include Kimi? Or Deepseek?
