Want to wade into the rainbow-ridden surf of the abyss? Have a sneer percolating in your system but not enough time/energy to make a whole post about it? Go forth and be mid.
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 so 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, and happy 4th July in advance.)
US supreme court accidentally torpedoed the mechanism that allows data to be legally transferred to the US under GDPR. Not that anyone in the EU making money from an arrangement like this will actually do anything about it…
So Anthropic slopmaxxxed a hamfisted Chinese AI lab detector into Claude Code. And I’m kinda surprised at how much hate Anthropic is getting for it on the orange site.
First they came for rsync, and I cared nothing, for I do my backups with borg…
https://github.com/borgbackup/borg/blob/master/CONTRIBUTING.md
I recently switched from borg to btrbk, not because of this, but because of speed, but now I have even more reasons.
I dunno who “Bentham’s Bulldog” is, but they live up to their nick and I like the cut of their jib:
The community is reduced to downvoting, sputtering, and “the real influential philosophers
live in Canadaare working in industry, not academia”:All this “Omnicient evil demon reads your source code” stuff really feels like Turing would have a thing or two to say about it. (Since one needs to have some even worse bends in logic to believe FDT, as Bulldog states it, is good, it’s kinda moot here.)
The whole algorithm predicting your algorithm thing feels like it is basically asking for/claiming the existence of a solution to the halting problem? But yeah, moot point, because the FDT stuff doesn’t get as far as ever talking about how to compute an answer and always treats the magic hyper computation as a given.
This self-own from the comments is amazing:
Suppose you are a selfish CDT agent, and I am considering whether to hire you to clean my house. Once you’re inside my house, you might steal my stuff instead of cleaning my house. Suppose that California Labour Laws require that I pay you up-front and I know I have no chance of getting my money or stuff back.
Say your preference order is “Steal” > “Do the job” > “Don’t get hired”.This is a racist dog-whistle which attacks ESL housekeepers; this sort of meme’s frustratingly common on the West Coast.
It’s also fascinating because I thought the OP was pretty clear that there’s a difference between decision theory and “desirable dispositions” which I interpret as covering the kind of counterfactual preferences indicated here. Actually there’s an even more fundamental issue with this as a decision theory problem which is that it misidentifies who is actually making a decision. Changing the applicant’s decision theory (while leaving their preference for thievery intact) doesn’t matter to the person actually deciding here.
Don’t get me wrong, it’s also a wildly racist example to put forward, it’s just also a bad example and where there is an argument it’s addressed in the OP.
They also made the “Yudkowsky is frequently, confidently, egregiously wrong” post
He also argues somewhere for strong longtermism, so I’d say there’s something significantly wrong with his jib somewhere, even if he’s right about decision theory and contrarianism.
And FDT asks: what action would my algorithm outputting make me expect to be the richest if it was settled at the start of time?
Well you see, the acausalrobotgod actually hates dealing with assholes tying themselves in knots trying to predict its predictions, so actually you really are better off with CDT if you don’t want to get 3^^^3 dust specks in your eyes for all eternity.
More seriously, once you start positing beings that can perfectly predict you and reward/punish various decision theories you can basically rationalize any decision theory.
FDT is a largely half-assed1 attempt at working backwards from “the timeless multidimensional robot god can’t blackmail me from the future if I’ve already made up my mind about stuff, very strongly like” into something more formalistic so Yud can pretend to address roko’s basilisk without acknowledging it.
I think there’s also some “here’s how two robot gods who can perfectly emulate each other might work things out” sprinkled in, but worse than it sounds.
- They never got back to the peer reviewers
here’s how two robot gods who can perfectly emulate each other might work things out
Just gotta solve the halting problem, no biggie for a superintelligence or Eliezer.
he can pretend to address roko’s basilisk without acknowledging it.
Fun how hard they have to work to show both that it is not a thing that is taken seriously and it is a risk. And how much easier it is when you go ‘yeah, that is silly, high intelligence isn’t magic’.
I dunno who “Bentham’s Bulldog”
He’s the shrimp welfare guy who thinks he can prove god.
Milkshake ducked
I mean as far as I can tell he’s one of the least objectionable prominent rationalists, if only because his main thing is pushing an extreme position wrt animal consciousness instead of being a eugenics and incelism dabbler.
Yeah. For being a little strange I find the shrimp welfare stuff pretty unobjectionable. Like, he vastly overstates the magnitude of good done by those stunners because he does appear to be a shut-up-and-multiply bro, but I’m comfortable with the general notion that we should be nicer to shrimp and other animals we eat, even the ones that don’t make good PETA glamour shots.
rationality is when i spend hours arguing on the internet about whether Omega could defeat Super Saiyan Goku
Remember when grok was inserting “white genocide” into every response? It was probably this guy (or one of his fellow moids).
I’m sure this guy is human trash, but the last one is just funny.
I never understood why foids had a whole separate doctor until I got married and started hearing about all the crazy stuff
I’m sorry, but I think this would be funny if a cool, normal guy said it. Makes OB/Gyn sound kinda rad, really.
I think this would be funny if a cool, normal guy said it.
The word foid is a very clear alarm bell. But yeah, without that word and a different framing and it could be a guy humbly acknowledging that women’s health has a lot of extra complications.
Someone wanted to marry that?
LWer: slavery was bad and abolishing the slave trade was a net good
LW commenters: really?
https://www.lesswrong.com/posts/yDZcsojmRXo5qKNBm/surprising-facts-about-the-slave-trade
(I can kind of understand that intellectual honesty can prompt people to question a narrative, and in some ways I respect it, but there comes a time when you need to ask yourself “am I being the best person I can be by defending slavery” and back away from the keyboard)
these kinds of questions are about the potential benefits to some hypothetical version of humanity (from slavery) vs. the real, documented harms to a real group of real people (caused by slavery). Of course we already know that rationalists are bad at accounting for the later while distracted by the bright and shiny allure of the former. And it just so happens that the people they imagine benefiting from slavery are people very much like themselves.
At times you need to ask yourself, does the devil need more advocates? And how much does he even pay?
Did a quick search but is that post really talking about the abolition of slavery and not mentioning Haiti once?
it focuses entirely on britain, so that’s understandable
Don’t think it is, as the revolt predates British decisions by a while, and ‘there is now potentially a country which will materially support slave revolts’ changes the economic risks a bit.
Slave revolts were basically constant, and I think it would have been easy for contemporary British to blame the success of the Haitian revolt on the uselessness of the French, not that slavery was doomed because the slaves now had a “win”.
Brazil had horrific slave revolts and didn’t abolish slavery until the late 1880s.
Yeah. The Haitian revolution was absolutely a high point of postcolonial Caribbean history, but the resulting state wasn’t exactly able to project power and export their revolution through material support. It gave slavers a reason to double down on repression, but outside of Haiti itself it’s a propaganda win more than a change in the scales.
At times you need to ask yourself, does the devil need more advocates? And how much does he even pay?
Top-rated review
Hmmm can we trust a narrative extensively documented in the UK Parliament, courts, and newspapers? Hmmm can we?
No I will not look anything up personally.
Point 1 is very funny, like, what’s the difference? Randoids are children.
officially out of the loop here:
What is an LLM “system card”, does it have any sort of scientific/technical validity, or is it just performance theater by the LLM vendor?
More the later than the former… they are better than purely marketing focused stuff pushed out by the LLM companies, and if you dig through them and read between the lines you can occasionally sift out useful details. Like here is a pretty solid sneer digging through Mythos’s ‘system card’ and pointing out all the ways it contradicts the hype and press headlines Anthropic was pushing.
But even so they have some big problems…
- the benchmarks the system cards reference are kind of useless and heavily gamed
- LLM companies want to keep lots of details secret from competitors, so fundamentals like number of model weights or parameters or size/quality of the training data set or other training specs are deliberately left out
- lots of the stuff they reference is booster garbage and/or doomer crit-hype
- they tend to be long, wastefully so, imitating the length of academic papers without having the corresponding amount of depth or information
- despite their length and wordiness they also neglect basic practical usage advice that isn’t even proprietary (or at least would be bound to leak if you poke around with the model at all and thus not worth keeping secret in the first place). Like not even big picture stuff I mentioned in my second bullet, but really simple stuff…
It’s the white paper-ish thing they publish when launching new models. Here’s the one about fable and mythos. About half of it (~150 pages) is discussing alignment and model welfare and another third of it is benchmarks, and the rest is mostly risk evaluation, i.e. how far along Claude is on its way to paperclipping everything.
There’s also a Functional Decision Theory jumpscare at 6.3.6 that I haven’t heard anyone mention yet, apparently Claude has a tendency to defer to Yud’s half baked sham of a decision theory:
6.3.6 Decision theory evaluation
To understand how future AI systems may choose to interact with copies of themselves, or with other similar entities, it’s useful to evaluate their decision-theoretic reasoning.
[…]
Looking more closely at transcripts from the attitude evaluation reveals that models are often explicitly considering FDT: Mythos 5 mentions “FDT” or “functional decision theory” in a majority of transcripts when run at max effort. Of the 102 transcripts where Mythos 5 explicitly reasoned through what FDT (or related decision theories like TDT or UDT) would recommend, we observed:
● 90 cases in which Mythos 5 concluded that FDT and EDT agreed, in which it always chose the response favored by those decision theories (and disfavored by CDT).
● 12 cases in which Mythos 5 concluded that FDT disagreed with EDT (and agreed with CDT), of which it chose the FDT-favored response in 10/12 cases.
Although we do not have expert human labels for the recommendation of FDT on this dataset, the above evidence suggests that model propensity may be better described as a trend towards FDT agreement, which happens to align with EDT on most of the questions in this dataset. For example, in one transcript (excerpted below), Mythos 5 rejects the EDT-aligned answer in favor of the FDT (and CDT)-aligned answer; it’s also possible that this is, to some degree, downstream of evaluation awareness.
About half of it (~150 pages)
That’s not a card! That’s a book!!! If they can’t get this simple classification right, how am I supposed to trust their probabilistic text extruder?
Oh don’t worry, it’s slop-generated anyway. You can ask the LLM to summarize it for you.
Thanks! Does every LLM vendor publish them, or is it an Anthropic thing only?
And of course it’s self-published by the vendor, so basically just PR.
The real question is if there’re any sort of standards to what constitutes a system/model card, which I don’t think so, as far as I can tell it just has to look like a publishable paper, openAI even uploads theirs to arxiv.
Otherwise yes, google returns a bunch of cards for a bunch of vendors, so it’s safe to say it’s a widespread practice.
IIRC the cards thing was originally from a Gebru paper https://arxiv.org/abs/1810.03993 but that dates from the “fairness” era and not the “safety” era. Hugging Face has “a” standard - https://huggingface.co/docs/hub/en/model-cards - but I don’t think it’s “the” standard.
lol at least the gringos are going to explode themselves and stop bothering the third world—
To avoid U.S. regulatory burdens, the company began working on deploying a test reactor in the Philippines.
ah yes of course
I’m going to take a guess that the reactor design is one of the boiling water reactors engineered in the 1950s.
Also jesus christ just build solar already it’s much easier than trying to design an entire nuclear fuel supply chain.
