HN user
A zigzag merge gesture is obviously a terrible idea until/unless everything is a touch screen. Did they even think about this stuff at all? Ergonomics and RSI aside, if a horizontal drag means add, why not just make vertical drag mean merge. Not a fan of voice interaction generally, but it's something we'll all be grateful for as we get older. No need to accelerate it
Yeah, but there's a point here. Are you comfortable with financial derivatives? Derivatives of derivatives? Futures of futures at the 4th, 5th or 50th order? The point being that if you go too far from the substance of things then you've lost the plot.
Engineers, especially SWEs, have lots of aphorisms to discourage exactly this and try to put it into professional doctrine and culture. (YAGNI, KISS, secondary-systems syndrome, etc)
Most people in management, finance, politics etc won't ever see it as bad unless they actually receive bad feedback. But bad feedback never comes if the incentive structures are broken (that's the point of TFA)
how do we actually get this implemented?
Hackers might be interested to know that there's an "open questions" section at the end of TFA. Some of it probably wants simulation, some wants theorems.
Camel-ai pubs/frameworks might be related and useful, for example: https://github.com/camel-ai/agent-trust
Several model checkers also have primitives for working with common-knowledge. TFA puts it like this:
Learning a fact changes what you know. Seeing it displayed publicly — where everyone else can see it too — where you know others can also see it, changes what everyone knows, and subsequently how they act.
An important piece of technical vocabulary, it really seems we need this to talk about a lot of problems lately. Here is Terence Tao talking about some related math for disinformation and politics ( https://mathstodon.xyz/@tao/114866548969775485 ) and summing it up this way:
we barely even have the vocabulary to discuss, let alone analyze, games in which control of information is a major battleground.
He kinda means in general though I think.. probably we can find heuristics and crunch a case or two
Or post it as a reply in url, hope for the silent majority to support it enough to send it to the top. If the platform only supports out-links it's tough.. maybe use QR codes as avatars?
This is amazing analysis, presentation, and has a call to action at the end. Some of this guys other stuff: https://tobias.cc/reading
The only point I'd add is that it's not handling time evolution in wicked problems quite right. Agree that the noisy room is distorting the world in exactly the ways described. But what if we've been in there so long, and the world has become so distorted.. that reality itself slides towards the once-extreme positions? Easiest to see this with climate-change controversy since that is the way that sort of thing happens, regardless of whether you think it's happened yet. Cascade, phase change, and collapse don't just call a truce.
So you have to anticipate that, acknowledging the pessimist is actually right, and that systems are a real bitch. Then you point out that if we're already doomed, we have nothing to lose nothing by trying. Systems are complex after all, that's the whole problem.. so if we miscalculated on the doom, then bothering to try actually saves us. Checkmate pessimists.
Pablo's group because when it was initially followed the leader was a Pablo, who was soon deposed by a chap named Cantsbee
Heh, I assumed Pablo was the scientist studying it, looks like we're in Swahili names now. That chap Cantsbee and the others all have wiki bios![0] Anyway Cantsbee does have a ridiculous and whimsical name that cannot compare to the honor and majesty of something like Ubwuzu, but he did ok for himself and his people, he really did.
a calm but powerful leader, rarely getting into altercations or fights with other gorillas. He led with grace, strength, and serenity. Cantsbee "resolved conflicts rather than starting them, protected his family with vigilance, and rarely resorted to aggression
Remember the Cant.
My own reason is .. maybe not possible to hit on research papers.
I think fancy people with appropriate credentials and .edu emails are all using openreview? So the audience is what, the unwashed masses who also happen to be doing some light reading at the bleeding edge of knowledge? Surely there are dozens of us I tell you, dozens! =P But yeah, maybe not enough to sustain a social network.
Never heard of alphaxiv, will try. I would also love for this to work, probably not willing to risk slogging through science twitter/bluesky/mastodon. Honestly HN would be the obvious place if it would add a pretty simple tagging system as most of the people interested are probably already here. I don't think we'll see that, because if we had filters no one would go to the front page, and that'd be a bad thing for certain interests.
Not sure you can purely talk about "is the motivation likely?" and end up with qanon stuff. This leaves out motivated reasoning coming from the rube, plus a bunch of other things like narratives that are sufficiently fun / scandalous /surprising
It's interesting right? Now there's too much distrust of authority and also not enough. Even the word "skeptic" is sometimes used to refer to people who "do their own research" and doggedly latch on to wild conspiracy theories.
Avoiding groupthink is another slightly different positive spin on (my read of) the underlying message. There's such a thing as toxic individualism too, but if there's a "bad" way to be a free-thinker then you could say it usually has a pretty limited blast radius for society in general and it isn't a contagious kind of madness either
Data here: https://gertlabs.com/rankings?mode=agentic_coding
Oh wow, we got "tribal domination", "market simulator" and "adversarial customer service". I don't know what those are but it sure sounds like big torment nexus milestones
Maybe we could at least play nicer games like hackenbush and act surprised when there's some wicked use-case that's isomorphic.
EDIT: Ok fine. I like "Rubik's Cube Chess" a lot. Never heard of it, is this analyzed formally at all? Hard to search for since there's tons of collisions
Oh this is a very damning paper. Using simple languages from their definitions alone is a great proxy for studying truly out-of-distribution reasoning. Also just for following simple rules/instructions correctly, because a simple enough language is practically just a grammar. This paper is terrible for anyone who wants to make the case that models can do those things well.
To the extent today's AI can reason, add this to the pile of evidence that you definitely need a harness. Counter to what you hear.. that seems true for SOTA and frontier, not just toy models. Lots of people were saying many years ago someone should test exactly this, because it's obvious. Someone at megacorp probably did try and decided not to publish because they thought it was bad optics.
Watch it if you haven't already. I accidentally landed in the middle of it while doing some illicit late night channel surfing when I was a kid.. this left quite an impression.
I think it was a healthy formative influence for me and primed me for rejecting fads / peer pressure, distrusting authority, etc. Probably also helped me to resist the more unhealthy aspects of a religious time/place, and I was even doing light reading on Cartesian skepticism a few years later, which got me into math. Didn't figure out the name of the movie until years later when it was a big meme.
This is not advice but I definitely advise you to show your small children this movie before they are old enough to think it's corny. They may have a schizophrenic episode or descend into solipsism sure, but they may also get scared as hell by monsters and learn some mental judo, and thank you for it later.
The Claude Platform on AWS .. giving you all native Claude API features .. Anthropic operates the service and data is processed outside the AWS boundary. This is a good option for companies that want the full Claude Platform experience.
Does seem to be mostly about billing like others said. But it might mean cloudformation / terraform providers for claude-platform, guess that's nice.
It might make strict networking/firewall things slightly easier somehow. But for everyone who thinks the new offering is about jurisdictional matters, it's not, that's the old one:
Claude on Amazon Bedrock keeps AWS as the data processor and operates within the AWS boundary. This is a good fit for companies that have strict regional data residency requirements or need their data processed exclusively within AWS's infrastructure.
I don't care if the app is a synchronous multi-page form with zero no need for websockets. It must have them
Sounds exactly like the kind of intuition an LLM will have.. "best practice" that's really whatever fads/marketing hype that there is a lot of noise about, never been informed by experiments or pain.
There was a post complaining about AI preference for god objects earlier, but the thing about stuff like that is, you could mechanically disincentivize it purely from complexity metrics or ASTs, either in training, or at the agentic layer later. I'm really much more worried about when LLMs are flooding the internet with marketing, and LLMs are consuming the marketing to determine best practice
Part of the practical degradation of traditional programmers over time has always been concentration and deep calculation, just like in chess.
Fortunately SWEs have the architect path, which frequently rewards having lots of deep intuition even as the details of calculation continue to change. So one question that's urgent but unknowable until we get there is.. are we going to get good architects if they don't come up through the trenches? I'm not sure. All I know is that everyone has a war story about the least qualified ones that got the role without that experience.
Since intuition is what LLMs do more than calculation, it's worth mentioning that this is true but different. They have the collective unconscious of the internet, which isn't taste that comes from good/bad experience. Besides intuition what comes to mind is "good taste".. the actual foundation of good review and really the main job of senior positions in any technical field.
I could write in brainfuck with ai, but I presume, wouldn’t get the same results than if going with python.
Insane naivety.
At the beginning, absolutely. At some point though if it keeps working.. a traveler that takes many risks might be better at evaluating them than a random person they could consult. And/or part of getting there is learning to evaluate which random persons to listen to ;)
Locals trend conservative, always giving the advice "don't do that, you'll definitely die" because they remember one bad incident 10 or 20 years ago and never clock how circumstances have changed. My favorite is the time I was warned over and over by several different people about not going somewhere because I would surely be killed by foreigners or wolves. Dude, what? If there was a major problem with one, it would scare off the other. Are they working together?? lol
I've got some demos of what the new Prompt API can do: Use surrounding context to rewrite your ad copy:
Yup, that's the plan. No local model, no webpage; more, better and cheaper adtech extortion/surveillance for vendors while everyone else pays for the juice and hardware degradation.
Probably you're focused on coding agents? I bet someone could use that kind of hardware to filter snarky comments
Entrenched interests are going to do everything to stop local, but there's at least a few technical reasons to believe small and specialized models could be the norm eventually. If that does happen, local will follow.
TFA is focused on whether big models are necessary for what users want. There's some evidence they may never actually be reliable enough unless a) mechanistic interpretation matures far enough or b) our multi-agent systems all become multi-model.
For (a), advancement in MI might fix problems with big models, but would also mean we can maybe get unified representations, and just slice and dice the useful stuff out of huge models, getting only what we need without the junk. Ability to isolate problems won't really come without bringing the ability to isolate functional subsystems. Only want logic? Only vision? Just cut it out of the big monster and enjoy reduced costs and surface area for problems.
For (b), just look at stuff like the evil vector, or the category of hallucinations specific to tool-use. Without a complete solution for helpful/honest/harmless alignment, it seems likely that creativity and rigor (and many other things) are fundamentally at odds. If you start to need many models for everything anyway, why do we need the huge expensive do-everything ones? So specialization also becomes a pressure to shrink everything towards minimal reliable experts
I'm grateful for the reference as context, but it simply does not settle the issue of transparency and it did reinforce the question. That's not controversial, nor is the statement that OpenAI avoids transparency, wants good publicity, will go to great lengths on both of these things like all corporations. It's also completely normal to think progress on Erdos problems is fascinating and inspirational, but aim for clarity on what the scope of the achievement really is (invention, composition, literature search) per the limited information available.
How does any of this offend a reasonable person? It doesn't, because here is the project wiki addressing BOTH of these relevant points. https://github.com/teorth/erdosproblems/wiki/How-have-AI-com... https://github.com/teorth/erdosproblems/wiki/AI-contribution...
What I raised is a real question: the erdos folks are not affiliated with AI companies.. but is the AI company affiliated with them? Actually knowing the user/org accounts involved is optional because they could just divert resources to anyone prompting near the problem. Anecdotally.. I've noticed what appears to be token-discounts based on topic, for example more generosity for AI-related research than random stuff, but it's hard to know for sure. Wouldn't you promote interactions you could profitably train on?
So again, allocating resources to Erdos one way or another is just a clearly smart business decision for something people are talking about and which has become an unofficial competition among vendors, not a big scandalous accusation. Something like a reasoning-trace is the only way to settle it. This isn't conspiracy or nitpicking because this is the topic itself: the AI usage is more a matter of public interest than the actual problem solution. What's the argument against more transparency?
I'm cautiously optimistic that the theory camp will benefit as much the applied guys, but later. They dream and scheme and shape fields by making vague intuitions and questions sharper, but they also need interesting patterns to work with as their raw material.
Identifying suitable problems in this sense rather than solutions is an AI use-case you don't hear about much. We don't quite have the infrastructure for this yet but by combining language models / anomaly detection / knowledge-bases we might be in a position to give a Conway 25 interesting high-quality puzzles before breakfast. Funny that it's like a kids nightmare, chatgpt giving them homework instead of solving it, but if it had good taste, people would love it for research.
Anyway, for now, dreamers will probably find more inspiration by cross-pollination with colleagues from different disciplines, or just going for a walk.
The duo had jump-started the AI-for-Erdős craze late last year by prompting a free version of ChatGPT with open problems chosen at random from the Erdős problems website. (An AI researcher subsequently gifted them each a ChatGPT Pro subscription to encourage their “vibe mathing.”)
Wonder who the AI researcher worked for? Is a "craze" something which a for-profit company would want to encourage? Maybe they'd think the publicity would help keep people talking about their company and product as we are now doing?
“There was kind of a standard sequence of moves that everyone who worked on the problem previously started by doing,” Tao says. The LLM took an entirely different route, using a formula that was well known in related parts of math, but which no one had thought to apply to this type of question.
Yep, I do remember this now. Everyone was yelling that this was definitely a sign of ground-breaking and creative work, citing the expert. What the expert actually said suggests that the solution was available in training data! That also suggests the math in TFA is harder in comparison, answering my other question.
Pleasure as always HN, thanks for voting me to the bottom of the thread for this
To get the perspective we need to see the nature of this anti-strategic self-slaughter, it will help to consider thirteen traditional bases of state power.
I've been thinking a lot lately about how the details are way past overwhelming now, that we need a framework to organize the sheer volume of stuff that's breaking down. This one works pretty well.. never seen it presented with this level of clarity.
We know this because most of the Erdos proofs made by AI have been done by amateurs prompting GPT 5.* Pro
Where's that? The stuff I've seen is from celebrities. Were those problems as hard as this one, or the ones that Tao posts about? Regardless.. what's the argument against more transparency here to just settle this kind of thing?
which is too conspiratorial of an explanation for my liking regardless
OpenAI is not, in fact, open. Why do they deserve the benefit of the doubt?
Regardless.. special treatment for special customers isn't conspiracy, it's SOP literally everywhere and especially if you're helping to beta test. Anyone who's ever interacted with any technical account manager has seen waived quotas, free resource allocations, etc. The quid-pro-quo is obviously that your cheap early access means you get to give talks at a conference (or make a blog post that a lot of people read and talk about).
Despite this coming from an independent expert and not from OpenAI, we need to be honest that this is more like a marketing campaign than open science. I assume the progress really is valid, the experts are indeed impressed, and we have the accurate time that it takes to produce the results. What we don't have is detail about true cost, or a CoT trace, or anything like that.
The implication is: we're ready to let everyone go wild with this very soon. Ok, go wild with what exactly? How do we know that influential VIP users who might make very friendly blog posts aren't getting allocated exclusive access to a billion dollars worth of hardware when they ask questions? I mean literally giving a certain group of people temporary privileged access to like 90% of all available compute would be a completely reasonable business decision for OpenAI.
Would a reveal like that change how we think about the result? What if half that amount of cash/compute could enable some completely non-AI approach of numerical brute forcing that settles the question even if it didn't write the paper?
My other question is always whether the latest is purely using giant models or if we're now deeply into harnesses that use MCTS and such. Understandable to keep that a trade secret I guess. But IMHO we should at least get the CoT trace as a proxy for true cost, or else maybe we're just getting played to do the hype for corporate.
I don't necessarily think engaging in the personalities is interesting, but I'm struggling to see what is the beef here. Is it personalities? Pure vs applied math? Or AI?
Now, if these to-be researcher students feel defeated (enough to quit), or completely lean on AI models the work for them, we're going to have a problem. [..] If we look at it a bit cynically: Some researcher will be able to pump out a lot more papers by spending money on compute, than a couple of years of training students.
Obviously this is already happening and will accelerate. Outside of grad work, you could already just buy a degree. Certainly in the softer disciplines, you can currently just buy a phd thesis and a good publication history. If you're in industry instead of academics, you can even buy a promotion. If your employer gives an AI budget to all workers then you quietly double that budget out of your own pocket for as long as it takes to get a promotion, then stop and just enjoy a bigger paycheck.
A very interesting comment from Baez, I'll just quote part of it.
Where does the value of thinking and having deep ideas come from? We need to think about this now. If it comes primarily from their scarcity – the fact that having certain ideas is hard – then indeed this value may drop precipitously when the manufacture of ideas can be automated. But if the value comes from the utility of the ideas – the benefit that the idea brings – then the story changes: perhaps creating more good ideas is actually better, not worse. Here I’m using “utility” in a broad sense, not just in the sense of what people often call applied mathematics.
In other words, mathematicians may need to adjust to a transformation from a scarcity economy to an abundance economy.
https://gowers.wordpress.com/2026/05/08/a-recent-experience-...