domain:philippelemoine.com
confuciuscorndog
17m ago
Weird. The person I asked pasted to me what's in my post like 10 seconds after I pasted them the prompt (and their first exposure to the riddle), so I highly doubt they made up all of the specific AI-styled formatting in that amount of time. They in fact then argued with me for a bit that it wasn't even the right answer, until I explained why they had misread it and it was. (They got confused and thought the AI had transported the chicken independently at one point.)
I do think these companies A/B test and do a lot of random stuff, so maybe for all we know their accounts "GPT-4o" is secretly GPT-5 or something.
MaiqTheTrue
Zensunni Wanderer
18m ago
But if the alternative is “Israel no longer exists, Jews get exiled again to face pogroms and potentially genocide,” the nuclear option is much more on the table and the consequences of it seem much less important.
And honestly I expect Israel to get more brutal, not less, as the world turns against them. A lot of the reason that Israel was willing to tolerate Palestinians and the chanting of “death to Israel” followed by rocket attacks is that Americans had their backs and they had access to American weapons. Now, there’s a move to basically treat Israel like South Africa to recognize the state of Palestine (which is less of a state than many American Indian reservations), and to divest and potentially sanction them. This backs them into a scenario in which they can no longer tolerate things that they would have before, and cannot assume that if something happens to them that they’d be allowed to respond. I think that world opinion on Palestine has made the response much more brutal than it would have been otherwise. This is their last chance to destroy the threat, and anything and anyone left is going to be untouchable in the future because the world won’t allow another invasion of “Palestinian” territory. So bomb the shit out of everything and hope you’ve given yourself a long enough head start to get ahead of the blowback.
With proxy wars like Hezbollah or other terrorist organizations, again, they aren’t given the right to invade to root out those things with conventional means, and the sponsors have spent millions to create these groups and arm them and give them intelligence, etc.
ControlsFreak
27m ago
Try to learn to enjoy the feeling of emptiness in your stomach.
Man, my wife 'taught' me that this is not necessary. She jams soooooo many vegetables into my stomach; I feel soooooo full on basically negligible calories.
Amadan
"I would put a screwdriver through your eyeballs if I could"
37m ago
And yet the funny thing is... (premium) GPT-4o also gets this (at least as tested) "raw", without any additional instructions warning it about a trick question.
I have a premium account. It did not get it right, even after I pointed out its errors three times.
I'm more interested in the questions that aren't worded in a tricky way that most humans would screw up. The ones where they corrupt a riddle to make it stupidly easy. For example, I saw one recently where they took the classic riddle, "Two mothers and two daughters order three drinks and each get one; how is this possible" riddle (where you're supposed to realize that if you have a chain of grandmother/mother/daughter, the one in the middle is both a mother and a daughter), and turned it into, "There are four women (sometimes with various emphases to really drive home that they are four distinct people), two are mothers and two daughters.... order four drinks...." At least some of the LLMs (I haven't followed differential outcomes from different LLMs) can't reason their way into saying, "This is a bullshit, trivial question." They parrot a 'reasoning' step that is, "One is both a mother and a daughter!" and somehow still bring it back to saying, "This is how two mothers and two daughters can have four drinks."
This is related to my interest in having an LLM with a "bullshit detector". The ability to actually think conceptually and tell me that some bullshit is afoot; that there's something conceptually weird about what it's seeing; that if we think properly about a thing, then it turns out to be kind of trivial. This is personally a capability that, without which, makes one of my major possible use cases worthless, but with it, would become incredible. That is, I have to read and digest a large number of academic papers. Frankly, due to all the screwed up academic incentives, I don't know if I'd say most, but at the very least many of them are essentially bullshit. Once I figure out what they're actually doing, what the core idea is, given my contextual knowledge of the rest of the field, I can conclude, "This is completely trivial if you already know about these other works," or sometimes even, "This is just wrong if you know about these problems." I can have that conversation with other humans who are reading the papers, too. "Do you think they're doing anything other than X?" "Nah; I think that's all it is." I need LLMs to be able to do this, but they can't even figure out that four women getting four drinks is a trivial problem, likely due to the fact that they're fitting a data set rather than doing conceptual reasoning. Similarly, we're not going to have a dataset that includes, "Here are the conceptual reasons why these various academic papers are trivial or bullshit." We're just going to have a dataset that includes all these papers parading how wonderful and novel and interesting these new developments are. (EDIT: Note that @confuciuscorndog says that if they anticipated corrupted riddles, then they could just create a dataset with a bunch of them and train on it. Maybe so, but again, I just can't see where we're going to get a dataset that can appropriately represent calling out bullshit in a bunch of papers.)
Dean
58m ago
There's also another avenue-
4. The nuclear states basically fedpost-spam the potential supplier and customer networks, so that no one know who wants to buy knows who a possible actual 'legitimate' supplier is and no one who could sell one knows who a 'legitimate' buyer is.
There are absolutely terrorist and extremist groups with high interest in WMDs and WMD-substitutes (we had the Tokyo nerve gas attacks, for one, but the field of bioterrorism is basically just weaponizing natural epidemics). However, the groups that have interest in obtaining nukes are not the same as the groups that have access to nukes, and so all you really need to do is break the chain of commerce between the client (the person who wants the bomb) and the supplier (the smuggler).
This can be done pretty effectively by just stirring doubt and distrust on both sides, especially as both sides are in a psuedo-prisoner's dilemma where both need to be hidden from the eyes of the government authorities to work. A terrorist group / proxy needs to believe they're not being approached by an agent of the government, but runs into the issue that the local government and the local nuclear handlers probably share the same appearance/accent/cultural mannerisms (because a corrupt supplier is part of the government). In reverse, the corrupt supplier needs to believe that not only is their potential contact not a member of the government as well (or an ally of their government), but that the sale won't reveal their otherwise hidden network. Both parties will be 'better off' if they can trust eachother and make the deal, but each also has major payoff incentives to 'defect' and not engage, for fear of revealing themselves.
This is why the more credible loose-nuke risks come in contexts of state collapse (where the state is no longer in a position to monitor/maintain control deterence), widespread corruption (where the ability of the state to monitor is compromised by the state's agents being routinely bribed), but especially black markets (where a standing economic exchange system exists absent, and despite, state efforts). These are the cases where there's more credibility on the suppliers as having access, and more trust on the buyers to getting away with it, and more reason for both to believe the other actor isn't part of the state.
ControlsFreak
1hr ago
"millions or billions of mosquito-sized drones"
The primary problem that needs to be solved here is the power problem, not an intelligence problem. Of course, perhaps there will be an AlphaBattery program along the lines of AlphaFold that focuses on trying to come up with new battery designs, but my kinda-looking-in-from-the-outside view is that the battery folks don't really have a shortage of ideas for new designs; it's the empirical testing work (and things like not having the device explode) that is really the long pole in the tent.
wifje
1hr ago
What are the reason against passing laws that would reward dating apps for every couple that marries as a result of them?
There are of course many ways that this could be abused, but i yhink especially because this would be directed at a large organization it would be easier to limit its abuse in a eay that for example rewarding couples wouldnt.
So why not? Wouldnt that solve onr of the large incentive misalignments that exist in the market currently?
Dean
1hr ago
Haven't all the easily-deployed-by-third-parties nuclear weapons been decommissioned?
All nuclear weapons, once obtained, are easily deployed by third parties.
A nuclear weapon is basically the payload (the bomb) and the delivery system (the missile), but the bomb itself is very easily deployable by third parties. For example, a B61 nuclear bomb, the primary gravity nuclear bomb maintained for NATO purposes, is less than 12 feet long and (~3.53 m) long, and weighs only 700 pounds (320 kg). This could easily fit inside a basic cargo truck or shipping container, especially if you cut off the unnecessary parts of the casing.
First-generation nuclear weapons (like the WW2 era weapons) were bulkier, but even the Little Boy used in Hiroshima was 'only' 9,700 lbs. That's not even a capacity cargo container track.
In short, if you can get the bomb, and get the bomb to a shipping container, you can deploy a nuclear weapon. If you can then get that shipping container contents onto a ship (or even just a boat), you can deploy it to any major port in the world.
There aren't backpack nukes with 4 digit arming codes written on the side in crayon any more.
You don't need the default arming codes to make use of a nuclear weapon. The idea that nuclear weapons become innert without the right code is basically Holywood and security theater.
All an arming code for a nuclear is, is the software password to use the pre-installed software. However, the nuclear weapon is fundamentally an analog device of 'conventional explosive to move catalyst to trigger nuclear chain reaction', and the software doesn't actually do anything past the triggering the conventional explosive point. (Rather, the software is about when the conventional explosive triggers, often by being tied to sensors for airmovement and altitidue that a ground-based device doesn't care about.) The bomb goes boom when the internal trigger explosion is triggered, regardless of what software is used or if any software is used. The 'you need the right password or the bomb goes innert' is really just the conventional-explosive trigger-control software borking itself and needing to be replaced. The bomb itself is still 'fine', the UI panel just isn't working.
Non-state actors, or even state actors who steal another side's bomb, just need to replace the software control system for the initial trigger, and controlled demolitions are an extremely basic technology in the civil engineering sector around the world. There are bomb designs where a jurry-rig trigger software may be less efficient- such as an implosion device that's not quite synchronous- but this isn't 'you don't get an explosion,' but rather 'the explosion is smaller than it could have been, but is still a nuclear explosion.' And land-based devices were always going to be smaller just due to being based on the land rather than airburst.
In short, all the arming code system really means for a loose nuke is that there's a period of time between when a deployable nuke is captured, and when it can be armed and trigger via replacement software. That could be days or weeks... but depending on how the nuke is obtained, it could be days or weeks before the state knows to start looking, or where, by which point a shipping container can possibly be on another continent.
orthoxerox
I realized I can change my flair color to Barbie pink
1hr ago
orthoxerox
I realized I can change my flair color to Barbie pink
1hr ago
They don't have enough nukes. And if they nuke Cairo or Mecca, they will lose the last of their remaining goodwill.
BaronVSS
Sort by controversial
1hr ago
Upvotes/downvotes are a rough numeric metric of how high status someone is in a given online community, which is the closest thing to "vibes" people experience in a real life community. From that perspective it is perfectly reasonable to care about your arbitrary internet points.
RenOS
Dadder than dad
2hr ago
AFAIK contemporary research has trouble actually showing advantages for the worst students, while there seem to be moderate negative effects for the best students. I have the impression that in a "strong" society, you can improve some of the worst performing groups by giving them help and good examples to follow while simultaneously harshly punishing, up to kicking out, troublemakers. On the other hand, if the troublemakers are not punished, they can drag down everyone so much that it overwhelms any advantage of exemplary behaviour or help from better students. But in the current climate this is not really investigable, so the research base is pretty bad, and the researchers are also far too biased to be trustworthy. There is also the "issue" that the current level of segregation isn't actually hard to overcome for a competent immigrant parent (in fact, highly educated immigrants basically end up in good schools by default without any effort, at least here), so the number of students that would improve in a better school is pretty low. Even low-education high-conscientiousness immigrants will leave bad schools quite fast.
Once upon a time, I'm sure faraway and more isolationist America seemed vastly less threatening to Europeans than the near enemies. In fact, this is exactly what American hegemony in Europe still depends on.
Once upon a time, America had never jumped into the sorts of ill-fated intervention to reshape the Middle East one might expect of other powers.
Once upon a time, Japan was once isolationist.
I don't understand why people act like China's historical inability to harm Europe or best it at the colonial game is some immutable part of its character.
MadMonzer
Temporarily embarrassed liberal elite
2hr ago
I was responding to both of you - I wanted to explicitly agree with you that Labour's coalition makes them likely to get housing right.
FaibleEstimeDeSoi
2hr ago
Yeah it's true and I'm writing this from one of those recently built single-family houses. But I think this development isn't major enough, will be limited because of the lack of infrastructure and can be stopped by various government policies. Like mandating different flat plans and sizes, reinvigorating public transit in smaller cities and making personal houses even more expensive tax wise. I'm not necessarily advocating for this, but that and many other things can be done to keep people in cities(despite the growing uselessness of living close to work when everything can be done distantly).
Stellula
2hr ago
But anyways, enough about how human speech works.
That is not how human speech works.
Intelligence is complicated
Indeed. “Is this thing intelligent” is probably the question, maybe a question that will never be answered. The Christians would say that this is because machines do not have souls and I agree with them.
sun_the_second
2hr ago
Is there a reason why Israel can't nuke the Arab League faster than they prevent them from doing so?
self_made_human
Kai su, teknon?
2hr ago
I must respectfully disagree, given that he's my own brother and I do know him.
We used to hit the gym together back in the day, and I was far more serious about it to boot. If he was addicted to lifting for the sake of lifting, I'd have known by now, but no, mf genuinely only does it to make other people seethe with jealousy. Sure, he probably does get some satisfaction out of starting to build muscle, but it's not the driving motive given how much he bitches after leg day.
sohois
3hr ago
Labour will naturally benefit from external events. Unless China invades Taiwan, we'll probably see peace deals in Ukraine and Israel, inflation will slowly but surely return to normal, the economy will gradually rebound into 1-2% growth, and the strikes which have plagued a lot of the country will magically clear up.
A strong criticism of the way the Tory party governed over the past decade is that they spent a huge amount of time tinkering at the margins, constantly passing new laws over the most meaningless things, and never really made any big changes aside from a Brexit that was forced on them. I expect Labour will be quite similar
orthoxerox
I realized I can change my flair color to Barbie pink
3hr ago
I doubt Israel's chances of survival will increase if they start lobbing nukes at Arab states. If anything, this might trigger a full-scale invasion of Israel by the whole Arab League to prevent them from detonating another one.
CertainlyWorse
Dedicated Pessimist
3hr ago
In my 20's I had my manager/mentor drag me to the gym after a particularly stressful week of projects/clients. He introduced me to weightlifting as he had a background in amateur sport for which it was a benefit. I've been lifting on and off (just bought a rack/barbell/bench for my home gym) and I love the mental and physical benefits. Like others I started doing it for the girls. Now I do it for me.
I was also introduced to running when I was 18 for work related reasons. I've kept up with that (on and off) even when a gym hasn't been available. I go through periods of no exercise and generally get fed up with lack of energy/brain fog/being overweight and then go on a health kick to get things under control.
From my point of view, its much easier to maintain fitness as a lifestyle (or at least something familiar that you can return to) when you are exposed to it while younger. Have you tried to introduce an adult to say.. swimming or bike riding if they've never been exposed to those activities? Watch their looks of incredulity at their suggestion. You might as well suggest belly dancing on a public stage.
Although if I was going to suggest something to someone who had never exercised, Hiking would probably be the gateway drug. Almost everyone can walk a fair distance and if you get them to do it in the sunshine there is the double whammy of Vitamin D exposure and endorphins. Combine that with being their 'exercise partner' to keep them accountable and you have a better chance of getting some traction.
oceanofsolaris
3hr ago
The handfull isn‘t that small (it’s roughly 5-10 years of progress in ML).
While that isn‘t a conclusive proof the trend will continue, it definitely does give some credence to it (I would give it maybe a 50% chance).
I think it’s a good idea to map out the future of „what if obvious trends continue for ten more years“, even if there is always a chance that things go worse or better.
oceanofsolaris
3hr ago
Who are some other of the many people you noticed?
I mean all kinds of people are on Dwarkesh‘s podcast, some certainly have achieved something (Zuckerberg) others maybe less (Aschenbrenner … though he has definitely done a lot with his so far fairly short adulthood).
But I am not sure I see a strong pattern here (unless your threshold for accomplishment is very high).
I kind of understand your point, but it seems to me that this is also kind of inherently a stepping stone thing; you're not going to get a LLM that can't go "Hey, wait a minute..." about the plate being on top of the banana instead of vice-versa but can dissect the fallacies of scientific papers. (Recognizing that the question is "wrong"/non-standard I think is also just fundamental to learning to call it "trivial bullshit". And indeed the modified fox/chicken/feed riddle is bullshit because the ability/requirement to take two entities across at a time trivializes the combinatorial/sequential puzzle aspect; you can solve it by taking any two combinations of entities across twice, so long as you bring one back for the second trip. Nothing about the feed eating the chicken etc. even matters anymore.)
If you're really interested in this, my understanding is that instruct datasets for finetuning are just a bunch of examples of the types of outputs you'd want an AI to copy, containing the lesson you want it to internalize, written out and then put in a particular format compatible with the base model. So if you could write a couple hundred of examples of your incisive smackdowns of scientific papers, perhaps formatted to some degree as a conversation between an assistant and a user, you'd be well on your way to that dataset.
I for one have dreamed of this myself, a "Steve Sailer AI", "Alexandros Marinos AI", or perhaps more generally "Crotchety Substack AI" that would be inherently skeptical of sources and do deep dives into them like the aforementioned writers, particularly as it regards searching on the Web, instead of just blindly spitting back at me whatever the top results or most "credible sources" say.
More options
Context Copy link