maxbond

If you're going to gamble, it's probably for the best that your counterparty doesn't also control the platform. I'm not saying that justifies being able to gamble frictionlessly, but it is marginally less exploitative. Eg, back in the day bucket shops (which sold binary options, like prediction markets do) would increase the spread in proportion to their assessment of your skill such that you would lose even if you were more skilled. In a proper market the platform makes the same amount of money whoever wins.

So, not all that different, but marginally less exploitative. I've never looked at Polymarket but Kalshi and PredictIt slid steadily from things of at least plausible real economic value (a market where it was conceivable [if unlikely] someone would be hedging their insurance contract or something) into total nonsense with no non-gambling function like whether someone would tweet a certain word.

I think prediction markets could serve a similar to a futures markets and have a functional purpose in the economy. It could be useful to generate real time estimates of the probability of some events that no one can control and have real economic consequences, like a hurricane. But evidently that's not where the money is.

JavaScript can be tricky.

Would you rather pay for a nonsensical explanation?

And also because it creates "one neat trick" where it can answer "I don't know" for many/most things and still get credit.

If you could write that reward function you wouldn't need an LLM, you'd just query the reward function to answer any question. You can create a benchmark and check that automatically, but you can't solve this in the general case. The model can do well on the benchmark but still give overconfident answers in areas the benchmark doesn't cover.

You can definitely tune a model to say "I don't know" more often but it will cost you performance, the model will reject some questions that it could answer meaningfully. In the degenerate case the model could collapse predicting that sequence always or almost always.

What's burned me before is iterating over hash maps. B-tree maps (or hash maps that are guaranteed to iterate in insertion order, or any fixed order) are your friend.

It definitely wasn't causation, the causation flows the other way (criminalizing a proxy for a political belief as an end run around the first amendment).

Yeah, I agree. It works as an argument for why science can't rely on wealthy benefactors, not for understanding what is possible with a given budget at a given point in time. And biology and chemistry are good counterexamples where capabilities are getting smaller and cheaper.

I would point out that that's on the back of a huge amount of research funded by grants and performed in national labs, but it doesn't impact your argument.

That sounds like excellent grounds for suspicious but I don't know what you mean. We were talking about Peter Higgs for example. I don't think Peter Higgs could have self funded CERN. I don't think a thousand Peter Higgs could have. Nation state level resources are the table stakes for fundamental research into particle physics, because everything beneath that barrier has already been explored - I don't think that's really controversial.

It's definitely an exaggeration to say that all science on a shoestring budget has already been accomplished, there are new frontiers out there. But once they start gaining momentum, the low hanging fruit will be consumed in due course. Methodically searching a domain works and works from the most tractable end up until it is at the frontier of what is tractable given our current technology/constraints.

I don't really understand the alternative hypothesis. That there's an infinite amount of low hanging fruit? What's this 100% failure rate?

What you're describing is (pre-)training. Distillation requires richer labels, the probability distribution over tokens (it would be logits rather than probabilities but that's not important). From a chat transcript you can only understand the argmax/most likely token of that distribution (and only if the API allows you to set the temperature to 0). It's not impossible for an API to give you that but they won't if they don't want you distilling their models.

The intuition is that distillation exploits not only the "right" answer but the relationship between answers (what's the second most right answer? the third? etc).

"Featureful" doesn't imply "feature complete". They appear to release minor versions all the time.

https://curl.se/docs/releases.html

If you dig into them you'll see there's lots of features that aren't adding new protocols. But incidentally they added a new protocol in March (mqtt). You'll also see that the list of bug fixes is prolific.

https://curl.se/ch/8.19.0.html

All of the low hanging fruit that could be discovered by self-funded gentlemen scientists has been picked. That doesn't scale to a supercollider or a large RCT. Funding at the whims of rich benefactors is very susceptible to petty politics.

Politics is irreducible from human affairs, privatization doesn't eliminate politics. It relocates it to a different set of actors. That could be a better set, but when it is it's because it's a more local and hands on group of people, not because those people happen not to work for the government. Governments are awkward because they are deep bureaucracies, and deep bureaucracies divorce the decision makers from the impact of their decisions. Weaker feedback leads to worse decision making. Not because there is a magic property of government that makes it uniquely bad. Large corporations, universities, and other deep non-governmental bureaucracies have similar pathologies.

That's something of an exaggeration, they are empowered to do violence and collect taxes and other things that are more problematic when abused, but still, privatization isn't a silver bullet.

That's a tree, but the rest of that comment is the forest.

Why are you interpreting clear communication of a window of downtime with 2 weeks notice as a "lack of cooperation"? That's what cooperation looks like. It's not explicit but my read was that they're not even taking a vacation - they're just doing the rest of their job, a lot of which is probably going to be shipping fixes for vulnerabilities that are already triaged.

To be honest I don't think my account is in 100% good standing, but I can't say for certain. There's definitely some dead comments on my account that are deserved and I think there are some small limitations that are or have been placed on it (probably fairly). Mostly around flagging and vouching.

Hmm. Interesting. If it was [dead], probably a false positive from a naughty comment filter; if it was [flagged][dead], difficult to say, potentially even an accident, or maybe people didn't like the joke. Given the non-negative karma, I would guess the first. Regardless, I appreciate the vouch.

I don't know what a next generation CAPTCHA should look like, but I know anything game-shaped will be a trivial target for RLVR. That's like trying to beat Stockfish. That ship has sailed.

It's massive and complex codebase. From the looks of it, pretty much what you'd expect, lots of chores, work on the test suite, keeping docs up to date, bug fixes. I didn't see any new features on my light skim but I'm sure they land occasionally.

https://github.com/curl/curl/commits?author=bagder

Atlas shrugged, but only for a month. I kid, it's well deserved. I do worry about their contract work loophole - if people disclose vulnerabilities publicly, their clients may pressure them to ship a fix anyway.

Again, while I respect that that is your view of sincerity, I don't agree, and I'm not talking about lying, which is obviously inappropriate. Framing is not the same as lying (though taken to an extreme it becomes a lie of omission). There's also just a misunderstanding of what is being asked by questions like "what's your biggest weakness?" Language can be ambiguous, just because a very literal and extreme interpretation of a question exists doesn't mean that is what is being asked.

I don't think we're going to bridge this gap but I also don't think that's really necessary. People have different values and it is what it is.

What I'm telling you is that it isn't an issue of integrity, and that only makes sense from a false premise - that the strictest, most blunt response is what is truest, what is being asked for, or a reflection of your alignment with the organization's values. That's really not the case. If I asked you how you were doing would you tell me about the traumas you're currently processing? Would you feel like it was a violation of your integrity if you didn't?

If that's what your values are, okay, I'm not going to tell you how to live, but it would be premised on a misunderstanding of what "hi, how are you today?" means.

I am not worried about what my pilot said in a job interview, I'm worried about what the check pilot thinks of their performance. Worrying about what they said in a job interview is like worrying about what they scored on the SAT. Once that hurdle is cleared, it instantly becomes irrelevant, because it was never measuring what we're actually interested in. It's a filter for people who are completely unqualified, it doesn't really measure a level of performance or alignment.

That's fucked up. If those are your values, that's all well and good, but you can't expect someone else to make the same decision.

Job interviews are a performance where you demonstrate you understand what professional expectations are and can abide by them. It's not dishonesty to not respond "I drink too much" when they ask "what's your biggest weakness?" just like it's not dishonesty to respond "can't complain" when someone asks "how are you today," even if you have a lot to complain about.

Once I interviewed someone and they described their tax fraud scheme to me. We didn't go with that candidate. Not per se because they committed tax fraud; because they demonstrated terrible judgment.

I don't think having trouble knowing how to tailor your message to your audience because of limited information implies it isn't truthful. Answers to jobb interview questions are usually very manicured and rehearsed but I don't think they're generally lies.

> Then someone points out that it has no anus and doesn’t need to breathe. But bodily functions? Harumph, why would a true duck need that?

Alternatively, your priors about ducks may be wrong.

What makes that more likely than that people at the DoD are alarmed (with or without good justification) at Fable's capabilities plus finding a jailbreak (or what they interpret as a jailbreak while Anthropic seems to dispute the requests met the level Fable should refuse)?

Sonnet 4.6 didn't have a problem responding to a prompt containing the first one. Some light searching surfaced a claim this stopped working very recently (May 2026). Perhaps related to the Fable rollout.

I like to say that every moderation primitive is a denial of service primitive and vice versa. ("Moderation" not being intended to imply it's good or legitimate. You can substitute "censorship" and it's the same statement.)

I think they meant more "it can be extrapolated or interpolated" or "it can be high variance and 'creative' or it can be low variance and 'reliable/correct/likely'". If you want to see something new, the model will need to step off the manifold. But the manifold is where you've learned the "correct" solutions live.

The ad-supported alternative suffers from the same principle-agent problem. What's to stop an ad-supported model from declining to refer you to products that would be better for your use case but who's vendors haven't paid the model's provider?

Ultimately if you can't trust the provider it is game over and you don't have an alternative other than to move to self hosted and open source solutions.

Is it possible someone on the team published a blog post about why it wasn't possible? Or that otherwise the AI learned from your 5 years of effort? Some of the code made it on GitHub or something?

It's totally plausible that it didn't & it still got it right, I'm just curious.