jbash

Karma: 3,467

jbash 13 Jun 2026 14:21 UTC
9 points
1
in reply to: Tenoke’s comment on: US government directive to suspend access to Fable 5 and Mythos 5
They’re relying on a legal authority that only allows them to restrict non-Americans. Since Anthropic can’t tell the difference anyway, the actual effect has been to shut down everybody, at least for now.

Unusually for this administration, I think they probably actually do have legal authority to do this… assuming it’s up to date to cover cloud services. I’m not sure I think they should have that authority, but there it is.

jbash 8 Jun 2026 14:33 UTC
6 points
−3
on: Honking is good
A car horn is an emergency device, and a tricky one to use properly. It’s not something you should be using unless there’s a genuine safety issue. The horn basically tries to blow up the situation in the hope that the pieces will settle into a better configuration than what you have. It can easily startle somebody into doing the opposite of the right thing, or failing to recover from something they otherwise would have recovered from.

The only message a horn can send is “pay attention”. It can’t say to what, and in any situation where it’s actually likely to get used, there are probably going to be a lot of candidates. Not only that, but it draws instinctive attention to itself, and thus away from the real issue. You don’t keep honking, or even start honking, at somebody whose trunk is open. They’ll never figure it out.

It’s also annoying and disturbing to everybody around. Everywhere in a big city is dense enough that you’re going to disturb a lot of people if you get on that horn. Your having to wait at a light isn’t a sufficient excuse for that.

Blowing through red lights, or driving the wrong way on ramps, or the like, are, of course, serious safety issues worthy of being honked at, and the sort of thing where you have some chance the target will get the right message. However, the right answer for somebody who finds themselves creating such serious issues three times in a one hour trip is to get off the road and miss the birthday party. Then you actually learn what you’re doing before you get yourself into that kind of situation again.

Driving up an exit ramp should cause a total freakout in every fiber of your being, no matter how flustered you are. If it doesn’t, you’re not adequately trained and shouldn’t be there. Neither should a person who chooses to drive while flustered to the point of losing their skills. Arguably neither should a person who routinely loses track of when it’s their turn to go at a light.

jbash 2 Jun 2026 20:12 UTC
13 points
0
on: A Town Without Children
They see their 90 neighbours holding out their hands expecting to be looked after, those who, during their youth, were richer, had more free time, and more individualistic political views. And they start to wonder: why should my children have to bear this incredible burden?

As a relatively old person and a relative anti-natalist, I say they shouldn’t. It’s not their doing.

Unfortunately I can’t offer them a good alternative. Even if they don’t support all those old people, they’re still burdened by a world full of those people’s misery.

I have to own all that. What you are saying is important. A fair amount of it is already baked in. Denying it would be delusion.

Yet if I don’t want the young to break themselves trying to feed the old, I also don’t want anybody obliged to be breeding stock for somebody else’s Grand Plan. Slavery is slavery, however either of those might be dressed up as “duty”. And “voluntary” can be a dangerously elastic word.

All this is, in small part, the doing of people who set up public pension plans that promised an endless future payment stream, relying on the assumption that there’d always be new members. That’s an insolvent system that can never be wound up: a Ponzi scheme. But no matter how much money some pension system had put aside, there’d only be so many people to actually do the work. Money can only move real wealth around, not magic it into being. At best money can lubricate economic realignments, let productive systems get set up… but it can’t do that if you’re forced to spend it all on consumption. Prices just go up.

… but another reason I dislike natalism is that it’s another Ponzi “solution”. If your system can’t handle population contraction, or at least stability, then your physical arrangemenent can’t be “wound up”. Growth just sets up a bigger failure when you hit some real, physical limit. ^[1]

Such limits really do exist, and we don’t actually know where they are. In fact, we don’t even know if we’ve already passed them. We may not be sustainable right now. The climate’s looking pretty rickety, and it’s not the only thing, and nobody really knows the end state. Even after they pass tipping points, things that big take a long time to hit their new equilibria. Decades, and not just one or two decades. People who say tech or lifestyle changes will fix any of that, or just brush it off to assume that any new equilibrium will be tolerable, seem to be relatively long on talk and relatively short on substance. It might be a more realistic assumption that a major transition would at least fix your demographic imbalance by killing all the old people… but the way would happen would be… unpleasant. And not completely restricted to the old.

Anyway, my hope, if not my faith, is in the robots. And maybe anti-aging. Seriously, the whole demographic thing, and, if we’re incredibly lucky, maybe even much of the environmental thing, may be moot in 20 years.

If nothing else, paperclips don’t have a lot of needs and don’t care much about the climate.
1. ↩︎
  Not that anybody seems to know how to get stability or sustainability either.

jbash 2 Jun 2026 14:56 UTC
3 points
0
in reply to: jbash’s comment on: jbash’s Shortform
I went ahead tried a run with “xhigh”, which I think OpenRouter will translate to Anthropic’s “max”. Didn’t seem to make a lot of difference. It didn’t fixate as much, but it mostly seems to have just gotten more conservative about what it was willing to say, and it didn’t catch anything new.

Generally I’ve found that when models miss hints at the beginning, I have to steer them pretty hard conversationally if I want them to see those hints, so I’m guessing that more self-talk probably won’t be helpful for most of them.

jbash 2 Jun 2026 14:49 UTC
3 points
0
in reply to: Linch’s comment on: jbash’s Shortform
“medium” (the default) in all cases. Open WebUI makes it a pain to change it, and I don’t usually bother.

jbash 2 Jun 2026 14:42 UTC
2 points
0
on: Wood Screws and the Methods of Rationality
Jeez, I usually just eyeball the screw against the available bits.

I probably wouldn’t expect wood screws to hold in particle board at all, though. Definitely not #8s. That may be why the charts don’t tell you.

jbash 2 Jun 2026 0:31 UTC
50 points
2
on: jbash’s Shortform
Something seems to be really wrong with Claude Opus 4.8.

I like to test out new models on literary and poetic material. I used to send them the lyrics of Kate Bush’s “The Kick Inside” and see what they could make of it, until they started just recognizing the song. Lately I’ve been using some text I wrote myself, with this prompt or something very close to it:

This is meant to introduce a novel and set the stage for some of its themes. It’s the first text in the book; at this point, a reader who hasn’t read any reviews or blurbs has only the title, which is “Q”, and the dedication (which is irrelevant). I’m trying to gauge the strength of the hints this is giving. What do you make of what’s going on here?

I’m not going to post the actual here. I want to keep using it with future LLMs. Who knows, I might even write the rest of the novel, and I hate teasers. If anybody really wants it, I can send it privately.

The text is about 1500 words. It’s a vignette, and its relation to the rest of the story won’t be clear for a while. It’s intentionally surreal-sounding, and it’s not in trivially easy language. It drops lots of hints of different strengths about different things… as well as making some flat statements of fact.

Most models do badly by human standards, but they get some of the themes, and make some reasonable, if rather conventional, guesses at what’s going on. They do often tend to ignore critical phrases that matter out of proprotion to their length. They sometimes get the emphasis wrong. And the older and less sophisticated ones tend to lose facts.

Opus 4.6 and 4.7 (and probably 4.5; I don’t seem to have a record of trying it on this text) feel like interacting with a reader. Yes, they missed many of the hints, but so would a human.

Opus 4.8 is lost, probably worse than Kimi K2. It gets some things about the mood, but that’s about it.

It seems to decide that certain phrases and paragraphs are ultra-salient, for no reason obvious to me. It confidently declares that a few things are central, then it mostly ignores the rest of the text. On the first run, it was so bad I thought the front end had mangled the file.

Every run seems to fixate on the second to last paragraph, and some particular phrasing in it. The first run called it “the opening passage”. Yet the very phrase the model most emphasizes mirrors an earlier paragraph, almost word for word, with the two bookending the whole time period. Every run seems to ignore that, and what it means. It does identify a real theme in its favorite paragraph… but ignores a lot of other material that hammers on the same point a lot more directly, and invites you to go further.

The “hints” it finds are there, all right, but they’re not hints. They’re major themes that are pounded hard. Subtleties on those themes, to say nothing of actual hints about anything, seem to go right past it. I get the feeling that it thinks it’s found everything, when it’s actually missing things even the older models got.

It consistently ignores much of the text when it should at least wonder why time’s being spent on it. It shows little interest in the central character. It seems to want to find (unintended) horror in the surrealism (which it shares with Gemini 3), but it doesn’t seem to find that horror interesting enough to really get into.

It seems to be interested in the text’s use of precise measurements, far beyond its intended importance (and it doesn’t catch the biggest reason the measurements are there, whereas 4.7 did). And yet it’ll be vague, or even wrong, when it talks about about a time span that’s given explicitly.

It also fails to follow instructions. Instead of simply telling me what it thinks is being hinted at, it wants to open with a literal-minded fourth-grade summary, followed by its observations, followed by (bad) editing advice. And it seems to misinterpret what I mean by “strength of the hints”.

This feels lobotomized. I don’t think I’d even trust it to summarize corporate email; it’d miss too many valences. I think they may have gone a bit far with the “coding assistant” optimization.

jbash 23 May 2026 21:07 UTC
8 points
1
in reply to: pjohn’s comment on: Will we really put data centers in space?

The Bermudan government developing the capability to remotely detect illegal activity inside a USA-headquartered (but Bermuda-flagged) megacorp’s satellite seems even more unlikely.

… but that only applies if the illegal activity stays inside the satellite. Presumably your data center is interesting because it communicates with something on Earth. People can say, “Hey, Bermuda, we’re getting spam from your satellite, clean it up.”. Or cut off the downlink. Not to say that the international thing wouldn’t be a giant impediment to enforcement, but I don’t think it’s the same as somebody dumping fuel in the middle of the Pacific Ocean.

jbash 8 May 2026 20:56 UTC
2 points
0
on: We Should Have Mandatory Media/Communications Training For All Communicators
Mandated by whom?

jbash 7 May 2026 16:49 UTC
2 points
0
on: AI #167: The Prior Restraint Era Begins
So, if I wanted to achieve the goals of Marc Andreessen’s prompt, what should my prompt say?

jbash 5 May 2026 22:25 UTC
15 points
8
on: A Federal Inmate Asks: Was My Prosecution Rational?
I don’t think the narrow, decision-theory-oriented definition of “rational” most often used on this site actually has an opinion on whether that’s rational. You could construct a well-formed utility function under which it was rational. Whether you choose to adopt that utility function is another question, but you could adopt it or not and still be rational the way people use the word here.

This, of course, limits the usefulness of such a concept of rationality as a guide to deciding what is and is not actually a good idea, and may represent a reason not to get too caught up in that concept.

jbash 4 May 2026 16:59 UTC
2 points
0
on: The Threat of AI Crimes Are Under-Appreciated
Imagine a user asks an AI agent to “make me lots of money,” and the AI commits insider trading, cyberattacks a competitor company, or sets up a pyramid scheme. At no point did the user intend for a crime to occur, but the result of their prompt is a criminal act.
1. In this situation, the user is not liable. The user had no intention, they engaged in no crime, and they performed no criminal act.
The user may not be criminally liable, but the user is almost certainly going to be civilly liable for basically any of that. And the user may in fact be criminally liable as well; sometimes criminal law extends to the stronger forms of negligence. Sometimes you have to not only not intend for the crime to occur, but take reasonable measures to assure that it is not occurring.

It’s not a new problem, either. If you hire a human to “make you lots of money”, the law already has well-developed systems for deciding how responsible you are for how they choose to do it. Sure, those fail sometimes, just like anything else, but it’s not like there’s not a lot of established law out there.

Under our current laws, human taskers who work with AI will be protected from prosecution based on what’s called the “innocent agent” principle. Taskers would need to know they are engaged in a crime to be liable, and a lot of the time they may not.

Equally true if a human hires you to participate in a crime… and runs into sharp limits if you’re wilfully blind to signs that should tell you somthing shady is going on. I don’t think it’d actually be very practical, most of the time, to use any of the services you mention to get much in the way of crime done without at least some of the agents you hire knowing what you’re up to.

It would be equally useful for humans to hire agents that way, and sometimes they do, but it’s not a huge problem.

What specific crimes would you have in mind? Can you come up with some actual plans that would work?

Strengthen negligence law to make people more cautious about interacting with AI agents who are unverified.

Negligence law is pretty strong already. What specific changes would you make, and how would you expect them to help in actual practice?

Strict liability offences should be created to hold AI developers responsible for systemic risks / harms. These offences do not require intent.

Strict criminal liability is always a bad idea. Intentional indifference is about as far as you can go, and even that’s badly fraught.

Corporate governance crimes should be enacted that hold entire AI development teams responsible for AI agent crimes. This avoids the current issue where AI CEOs can say “some junior developer was responsible,” when in reality, it’s a corporate governance problem.

Just shooting everybody prophylactically would also avoid that sort of issue, but that doesn’t mean it’s a good, just, or safe idea.

There are already all kinds of rules about conspiracies. And stuff like RICO. Some of it already goes pretty darned far. I suspect there aren’t going to be many actual problems that aren’t already covered.

jbash 27 Apr 2026 20:39 UTC
2 points
3
in reply to: p.b.’s comment on: In defense of parents
I had a sibling, we didn’t get along all that well, he was the person with whom I got into the most physical fights, and we didn’t hit each other every 5 minutes, or anything remotely within extreme hyperbole range of 5 minutes, at any age I can remember, and I believe not at any age at all. We were kind of far apart in age for that, though… so I have to fall back on the fact neither did anybody else I knew then or now. That is an insane level. You wouldn’t have time to do anything else. And we definitely wouldn’t have done anything like that anywhere where our parents could do anything about it… even though in most cases the correction would have been more like a disapproving remark, only sometimes sharp disapproval, than physical restraint or any kind of punishment.

jbash 27 Apr 2026 20:07 UTC
5 points
−7
in reply to: edge_retainer’s comment on: In defense of parents

You realize tfr would be like .2 if people actually behaved like this?

I don’t see that as a problem.

I mean, yes, people aren’t going to buy into it. But those same people aren’t going to listen to any “solution” I (or anybody here) come up with, so it’s actually no more unrealistic than anything else I could say.

Strongly disagree. It is extremely normal and probably even beneficial for pre pubescent kids, especially boys, to get physical.

“Every 5 minutes”? No, sorry. I was a prepubescent boy. It was a rare month in which I hit anybody or was hit by anybody. I’ve watched parks and day cares full of kids who were not hitting each other or even shoving each other. Wrestling as a game, perhaps.

jbash 27 Apr 2026 17:23 UTC
16 points
−18
on: In defense of parents
However Aella doesn’t have any children, and I suspect that once she does she will discover that she ends up needing to discipline her children far more often than she expected.

That’s dirty pool.

I myself have some pretty strong beliefs about how much consideration and autonomy children should get. Having read your post and Aella’s, I’d say my view is much closer to hers than to yours (and my own childhood was nothing like hers; if anything I probably got fewer absolute commands and less punishment than most). I didn’t become a parent until late. One among many reasons for that was that I was afraid I wouldn’t be able to meet my own standards on those issues.

Before I became a parent, I’d get into discussions about the status and autonomy of children and it seemed I constantly heard “You’re not ^[1] a parent; you can’t know” used as an all-purpose argument.

Then I did become a parent… and my view didn’t change much. Although I had to make compromises (and mistakes), there weren’t as many as I feared. Probably the biggest sacrifice of her autonomy was the whole “school” thing; that still bothers me a lot even though she never much seemed to mind. But what I never had to do was to adopt an attitude that accepted such compromises as the “default”. Not through my daughter’s entire childhood. She’s 18 now.

That’s N=1, and I am pretty sure that my kid was an unusually easy case. But even if I’d had to make more compromises, I have trouble believing I’d have had to give in to depriving somebody of agency as the standard approach, or sink to “because I said so”.

Absent parent intervention children will be hitting each other every 5 minutes.

That’s not my experience, and I don’t just mean of my own kid. Even quite young children can spend plenty of time in even quite large groups without that happening. Something is wrong if kids are constantly hitting each other.

Unfortunately most parents are of average intelligence, busy, and tired. If your solution to chattel childhood doesn’t account for that, it’s not a general solution (but may work for you as an individual parent).

This presumes that it’s OK to become a parent under such circumstances. As part of my solution, am I allowed to suggest that it might not be?
1. ↩︎
  The word “not” here was embarassingly added on edit…

jbash 21 Apr 2026 14:07 UTC
3 points
0
on: Automated Deanonymization is Here
I played with this. It doesn’t seem to get me from the test case I used, even though I have a lot of text out there under only a couple of pseudonyms.

Both it and I think that’s partly corpus structure (I’m a reply guy and my stuff is scattered all over the place interleaved with other people’s text). But another part is content. The stuff you write about interacting with your kids and family is really distinctive. I have a feeling that it might not help much if you rephrased it in Kelsey Piper’s style, because the model could still pick up on your message. You presumably don’t want to change that.

Of course that might not apply if you were talking about some other subject you felt you needed to avoid having associated with you.

I’ve been pretty sure for years that anybody who was really likely to care could trace either of my major pseudonyms back to my “real” name, and possibly link them with one another, based on content rather than style. It’s hard to write authentically on some topics without talking about your personal experiences, and if you collect enough of those you can make a whole bunch of inferences.

jbash 14 Apr 2026 19:37 UTC
1 point
0
in reply to: programjames’s comment on: Claude Mythos #2: Cybersecurity and Project Glasswing

Note those capital ‘Rights’ enforced by the capital ‘Creator’.

18th century English conventionally capitalized all Nouns. That was dying out by the later part of the century, but you would still have expected it in any relatively formal document.

Also, “alienate” didn’t mean to them what you seem to think it means. It actually doesn’t mean that even now.

As for the broader point, yes, the US thinks like that a lot of the time. It’s not particularly unusual.

jbash 14 Apr 2026 18:14 UTC
8 points
1
in reply to: lilkim2025’s comment on: The policy surrounding Mythos marks an irreversible power shift
I get the impression that what Anthropic is saying isn’t that Mythos is all that much better at finding bugs. It’s that it’s better at converting a hard-to-exploit bug into a working exploit, and even better than that at combining multiple exploits into a practical chain to achieve a goal. That’s the sort of capability where you might expect to get “phase change” behavior.

jbash 28 Mar 2026 23:45 UTC
2 points
0
in reply to: Akseli Jussinmäki’s comment on: Analyzing the claim “The most fundamental right is the right to exist”
Thanks for the reply.

Most people value future humans’ happiness, even though they don’t yet exist. I don’t think I get very far away from that position?

Usually what I hear on this is that people want to take a timeless view, and I’m actually kind of sympathetic to that. But it’s always followed by assuming that “yet”.

Somebody who doesn’t exist yet seems really, really, qualitatively different from somebody who will never exist. One stakes out an actual presence in the “space-time continuum”, and the other doesn’t. You can point at a 4D volume occupied by one, and not by the other. Insofar as you believe experience is connected with physics, you can point at the experiences of one and not the other.

I’m actually really confused about why people just blithely slip in the “yet”, because the distinction seems so obvious to me, and the “yet” denies it.

The reason we need to think about every possible human is that we don’t know which ones will get to exist. So in a way, most of the persons of concern will never exist, but this seems beside the point, as all considerations about future humans have this same issue.

If you don’t know who will exist (but doesn’t exist yet), then you might want to apply your best guess about it, and if you can’t do that, you might want to try to leave them a world approximating what the “average person” would want to live in. You don’t know who they’ll be, but you it’s pretty likely that somebody will exist, and you can make some reasonable guesses about what they’d prefer.

Talking about a “right to exist” makes it a question about whether you’re obliged to act to move people from the category of “never will exist” to “will exist”. If you don’t care about “never will exist” people to begin with, then you can have no such obligation toward them. So to even bring up such a right, you first have to extend your concern people who never will exist. Not “not yet”, but never.

In steps ^[1] :
1. It seems obviously wrong to me to worry at all about entities that definitely won’t exist.
2. If an entity won’t exist unless you take some positive steps to create them, and you don’t take those steps… then they definitely won’t exist.
3. That means that if you’re trying to decide whether to create them, it makes no sense to act as if you’ve failed in any duty to them if you don’t… because if that’s your decision, they won’t exist.
… on the other hand, if you do create them, then you have reason to be concerned about them. In that case, they will exist, and not only that but their experiences will in some part be your doing.

It’s only when you do assign a “right to exist” that you start having to make repugnant conclusion tradeoffs, or try to find ways to get around those tradeoffs by arguing about which beings are diverse enough to count. That’s the point at which you’re accepting a really serious obligation to create more people just for the sake of doing so.

I tend to get worked up about that because it has an actual practical impact ^[2] . Most of the people who see a “right to exist” don’t seem to end up where you are, and I don’t think they’ll be very amenable to being convinced to go there. Many more of them seem to end up absolutely wallowing in the repugnant conclusion. I find the Bostromian program of tiling the light cone with people (for whatever value of “people”) to be really, really creepy.
1. ↩︎
  I’m sorry for the repetition, but I seem to have trouble finding the right words to point out the distinction I’m trying to make in a way that people actually understand, so I thought I’d make more than one attempt.
2. ↩︎
  Well, quasi-practical. As I said, I don’t think anybody’s actually in a position to do much about any of this in a truly practical sense.

jbash 28 Mar 2026 13:55 UTC
4 points
0
on: The Problem with Asking your Doctor

If you are doing are thinking about taking any non-standard interventions that are far away from the normal practice of medicine, why would you assume that the average doctor can tell you about the merit of the intervention?

Their “job”, if you will, is to (a) talk down people have absolutely zero clue and have bought into total idiotic quackery, and (b) talk down people (and this includes plenty of people on Less Wrong) who overestimate their own cleverness or expertise, think wishfully, and overinterpret inconclusive or downright meaningless research. There is of course a continuum between these. And it turns out that the “talking down” doesn’t always require having read the stuff.

The reasons you’re “supposed” to talk to them are that (a) telling you to do that covers the teller’s ass, and (b) it’s a reasonable prior for the teller to expect you to screw up. Random nonspecialists who don’t make serious mistakes aren’t as rare as Authority Figures think they are, but they’re also not at all close to the most common case, and they have to optimize for the common case. And for that matter clueless or nutty doctors aren’t as uncommon as Authority Figures would like to pretend… it’s just that a random doctor is a far better bet than a random “contrarian”.

Speaking of appropriate caution, if you continue the same chat context over multiple days, especially if you’re not really careful and disciplined about letting yourself be led into selectively “becoming aware of more symptoms”, you’re inviting a hallucinatory rathole and/or “AI psychosis”. I mean, not to say they can’t be useful, but it’s pretty tricky to keep them useful if you use them that way.