639 karma · joined September 10, 2025
What? If we're discussing novel synthesis, and it's being contrasted with answer-from-search / answer-from-remix.. the problem does not matter. Only the answer and the originality of the approach. Connecting two fields that were not previously connected is novel, or applying a new kind of technique to an old problem. Recognizing that an unsolved problem is very much like a solved one is search / remix. So what happened here? Tao says it is
> is largely consistent with other recent demonstrations of AI using existing methods to resolve Erdos problem
Existing. Methods. Tao also says "This is a demonstration of the genuine increase in capability of these tools in recent months". This is the sentence everyone will focus on, so what is that capability?
> the more interesting capability revealed by these events is the ability to rapidly write and rewrite new versions of a text as needed, even if one was not the original author of the argument.
Rejoice! But rejoice for the right reasons, and about what actually happened. Style and voice transformations, interesting new capabilities for fuzzy search. Correct usage of external tools for heavy-lifting with symbolics. And yes, actual problem solving. Novel techniques, creativity, originality though? IDK, sounds kind of optimistic based on the detail here.
> If your job is only to write beautiful code, you have a problem.
Definitely no one at senior or even mid thinks this is what their job is about. Something like modularity is beautiful because it works and because of what it enables.. we don't try to shove it into place because it's beautiful. Talking about it the other way around sounds like a manager who does not understand engineering trying to paraphrase stuff that engineers are saying, poorly. Indeed quoting from this self-described entrepreneur's "About Me" page:
> While I was there, I took a couple computer science courses, decided I was terrible at it and swore to never write software again.
I guess that's thought-leadership for you.
Yeah this seems political, and it is, but it's really about cognitive bias. Reframing the thing in terms of daily workplace dynamics is pretty easy: just convert "legally consequential facts" to "technically consequential facts" and convert "cultural outlook" to "preferred tech-stack". Now you're in a planning and architecture meeting which is theoretically easier to conduct but where everyone is still working hard to confirm their bias.
How to "fix" this in other people / society at large is a difficult question, but in principle you can imagine decision-systems (like data-driven policies and a kind of double-blind experimental politics) that's starting to chip away at the problem. Even assuming that was a tractable approach with a feasible transition plan, there's another question. What to do in the meanwhile?
IOW, assuming the existence of citizens/co-workers that have more persistent non-situational goals and stable values that are fairly unbothered by "group commitments".. how should they participate in group dynamics that are still going to basically be dominated by tribalism? There's really only a few strategies, including stuff like "check out completely", "become a single issue voter", or "give up all other goals and dedicate your entire life to educating others". All options seem quite bad for individuals and the whole. If group-commitment is fundamentally problematic, maybe a way to recognize a "good" faction is by looking for one that is implicitly dedicated to eliminating itself as well as the rival factions.
You jest, but the best way is to give the nerds only a feed and don't allow any other organizing principles for navigating the content. Lured by the promise of something interesting sometimes, people who are trying to follow their curiosity for math/archaeology will stumble into threads about corporations and products, throw ideas out in public for free, get sucked into politics, etc etc
This is not really your average front-page at the moment, and in fact it feels significantly more geared towards curiosity and less towards commercial interests compared with the usual. Partly because it's the holiday slow down but you see the same effect on weekends, which is telling. And IMHO hacker news is consistently more curious when America is sleeping.
HN desperately needs tags, which would serve everyone well, whether you're really here for curiosity or for all the corporate news. I'd really like to be able filter out most of the stuff for "working professionals" even if it is still legitimately technical (for example the recent docker and github news). What I usually want is all the hobby tech / PLT / open-source / science / learning-oriented stuff that's left over after that filter. I can imagine various reasons why forcing a "feed" style front-page is advantageous to some people.. but let's just say it's not about promoting curiosity.
If someone is forcing you towards high stakes tight-coupling with no thought whatsoever towards the lock-in, you should get it in writing that "we at ${org} are fully committed to ${vendor} with ${platform}, on ${cloud} using ${tech} come what may, now and forever" and lots of sign off so that everyone knows who to blame when this is inevitably wrong.
Notice that if you go this route you get no additional safety, and actually introduced an extra hazard.. don't lean in or your goggles get caught in the machine and drag your neck closer to whirling blades! Sorry if it's rude, but there's no nice way to say that this is bad to do and even worse to advocate for. If you don't get it, ok, go be as unsafe as you want when it's your ass on the line. No need to take the next step towards also trying to kill safety culture.
A bidirectional bridge that spans multiple representations from informal spec to semiformal spec to code seems ideal. You change the most relevant layer that you're interested in and then see updates propagating semi-automatically to other layers. I'd say the jury is out on whether this uses extra tokens or saves them, but a few things we do know. Chain of code works better than chain of thought, and chain-of-spec seems like a simple generalization. Markdown-based planning and task-tracking agent workflows work better than just YOLOing one-shot changes everywhere, and so intermediate representations are useful.
It seems to me that you can't actually get rid of specs, right? So to shoot down the idea of productive cooperation between formal methods and LLM-style AI, one really must successfully argue that informal specs are inherently better than formal ones. Or even stronger: having only informal specs is better than having informal+formal.
I'm sure we've agreed on this before, but I agree again ;) There are dozens of us at least, dozens! There's also a recent uptick in posts with related ideas, for example this hit the front-page briefly ( https://news.ycombinator.com/item?id=46251667 ).
I was tempted to start with alloy/tla for my own experiments along these lines due to their popularity, but since the available training data is so minimal for everything in the space.. I went with something more obscure (MCMAS) just for access to "agents" as primitives in the model-checker.
Yeah, no. The whole "every perspective has some validity" thing won't really apply to most safety/security issues. The most charitable thing to say here is that the workflow is completely broken. Less charitable but also valid is pointing out that it's actively harmful, and deliberate. I would be really surprised if this hadn't ever caused serious consequences whether a whistle blower was fired, an abused spouse got extra abused, or an informant was killed. If you think you've got a "valid perspective" that prioritizes mere user-discovery over user-safety, then you should not be attempting work that's close to safety and security, full stop.
What leaked was that I was a signal user, and that the person on the other side was a signal user. The security implications are obvious, and by itself, that's already enough to get someone who really needs to care about privacy killed.
> Also, I think there is a setting in Signal to prevent that
False. It happened without my permission as soon as the app was installed, and there was no way to opt out. Maybe they changed it since then, but the fact remains they obviously cared more about network-effects and user-counts than user privacy.
Sigh, there's just no need for this kind of apologism. You could just admit that a) it's bad behavior, b) they did it on purpose, and c) it's not possible to trust someone who does something like this. I'm aware they are nonprofit, so I don't know why it's like this, but the answer is probably somewhere in the list of donors.
Should have deleted my account instead of just removing the app, because it turns out the difference between using signal and using SMS is obscured for most phones, and when people thought they were texting me they weren't. I was just out of contact for a long time as people kept sending me the wrong kind of messages. I suppose one could argue protecting contact/identity is not a real goal for e2e encryption, but what I see is a "privacy oriented" service that's clearly way too interested in bootstrapping a user base with network effects and shouldn't be trusted.
Almost every sentence of this piece is a very powerful reminder that we're not really talking about education vs cheating and it's actually about real work vs optics, appearances vs reality, fake news vs information, and all the rest at the same time. A certain amount of bullshit is and always has been standard, and you see it in all kinds of folk wisdom (e.g. "the people capable of being politicians are the least qualified", "those who do not steal steal from themselves", "the market can stay irrational longer than you can stay solvent"). But in a very short period of time, society itself has shifted away from rewarding real effort or real results almost everywhere.
I agree that game-theory is a pretty good way to understand it, but the conclusions are pretty dark. Defection as the only available strategy and equilibriums that add up to large-scale attractors that we maybe cannot escape.
Really the paper is about mechanistic interpretation and a few results that are maybe surprising. First, the input representation details (base) matters a lot. This is perhaps very disappointing if you liked the idea of "let the models work out the details, they see through the surface features to the very core of things". Second, learning was burst'y with discrete steps, not smooth improvement. This may or may not be surprising or disappointing.. it depends how well you think you can predict the stepping.
> One might think this means that imaginary numbers are just a mathematical game having nothing to do with the real world. From the viewpoint of positivist philosophy, however, one cannot determine what is real. All one can do is find which mathematical models describe the universe we live in. [1]
Assuming 10x is real, then again the question: why would anyone do that? The only answers I can come up with are that they cannot share it (incompetence) or that they don't want to (sabotage). You're saying the third option is.. people just like working 8 hours while this guy works 1? Seems unlikely. Even if that's not sabotaging coworkers it's still sabotaging the business
Maybe parent is a galaxy-brained genius, or.. maybe they are just leaving work early and creating a huge mess for coworkers who now must stay late. Hard to say. But someone who isn't interested in automating/encoding processes for their idiosyncratic workflows is a bad engineer, right? And someone who isn't interested in sharing productivity gains with coworkers is basically engaged in sabotage.
Typical example of a extraction/exploitation mentality where innovation would be better. Wolfram is in an amazingly good spot to spin up better "simulation as a service" if they would look at fine-tuning LLMs for compiling natural language (or academic papers) into mathematica semi-autonomously and very reliably. Mathworld is potentially a huge asset for that sort of thing too.
My next go-to test for this kind of thing would be converting bananas to petabytes, hoping the backend is smart enough to try and use the bekenstein bound. Wolfram (still) fails this kind of test at the moment.