perching_aix

Plain common sense. The same common sense that enabled people to discard the words of idiots on Reddit who claimed that ChatGPT was just Indians in a call center (a real talking point I ran into multiple times back then).

OpenAI does seem to be genuinely just high on their own supply for a little over a half of the manuscripts they released, but for the other half that is machine checked, either ((internal model)) managed to find a way to cheese 4 independent Lean checker kernels at the same time, and OpenAI somehow hasn't noticed / didn't report despite the heavy reputational risk, or the results are just plain correct. Go figure.

But even for the half that wasn't machine checked, there's an awfully big quiet regarding any defects in them. Given the insane schadenfreude potential and the highly potent David vs Goliat framing associated, one would expect any counterarguments to be frontpaged immediately, and be circlejerked to oblivion. Given that is yet to happen as well, go figure even harder.

i tend to run a pass of sol 6.1 at its output, cleans up the language well and finds defects opus missed

would be, yes... are we going to pretend that botch jobs don't exist, and that everyone credentialed actually does an equally (or at all a) good job or something?

you either trust and/or verify. the people clearly cannot verify and clearly do trust. vagueposting about ai bad won't break that vacuous trust, on the contrary, due to the incentives involved, induces distrust on the human expert side.

so maybe you wanna actually elaborate?

the article above is about a defect in google's brotli compression library (CVE-2026-111497). compression libraries don't really do web serving, or communicate with networked resources at all, they're hermetic. so exposure depends on the context this library is deployed in as a dependency.

put it in the network path, it can become an RCE.

put it in a privileged context, it can become privilege escalation.

put it in a hardened environment, it can become completely inert.

in no sane perspective is me listing off the applications shipping with vulnerable versions of this codec off-topic. just because the article is hellbent on portraying this as some sort of second php2shell, the actual underlying issue here is still CVE-2026-111497.

how will they tell that the person hired did a good job if they can't tell if the ai cocks up?

git for windows is vulnerable even in its latest release (2.56.0.windows.3), careful pulling repos over https from random ass servers (prolly uncommon but hey). they already merged the fix on the repo, it's just in a snapshot build only; should prolly arrive on Monday (2.56.0.windows.4).

also affects codex (gui), which ships its own git, could be more of a problem there, when the LLMs get quirky at night. still unlikely to be a problem though.

speaking of, update codex (and wipe its cache at ~/.cache/codex-runtimes), ships its own node. update python.

stuff like wireshark and imhex remain vulnerable, even on their latest repo commits, which may or may not be spicy.

also Microsoft 365 Copilot still has its own unpatched brotlidec.dll from Oct 7, if anyone uses that. no fix available for that one afaik, not sure about how it can be made to access the vulnerable decoder.

sounds like it's not just these models that could really use a thesaurus

Nohow, because the real world is informal. This is as far as coding can and will ever go, basically.

And like yes, it does just launder defects to a layer higher, but another way to phrase that in this case (and in many other cases) is that it raises the bar.

Instead of defects being mostly about nonsense minutia (relative to business logic), they are about actual design decisions and their correspondence to reality, i.e. the actual original goal and point of this god forsaken, navel-gazing field.

Definitely at best a "sort of" type perfection. The endgame here is formally verified everything, at which point the only question is whether the then-consistent model that's implemented is any reasonable in practice.

Think "this is not a vulnerability, as no security boundary is being unexpectedly crossed" except the way the security boundaries are laid down is utterly stupid and hilatiously nonsensical. Same for non-security related defects ("working as expected, wontfix" -> "the way it works is dogshit, and so your expectations are idiotic too").

Now, since these things are supposed to align with human sensibilities and are supposed to capture the human semantic understanding of our world, theoretically, this should still leave us in a pretty good spot. But not a guaranteed good spot, as that is not actually tractable, ever, and fundamentally so.

So when this will inevitably result in a paid web emerging with higher trust information for an additional fee, a fee that these companies will happily offload and/or partner away, will you be there to cash in as a mighty coincidence, or is this realization still awaiting you? Among the many other hilariously fucked realizations.

What's the paradox about this? It's just basic lying and deception.

On the contrary.

"weights open" == "you can download them", not "you know how they landed on them". that'd be an "open source" model instead, as they'd have to provide both the training data and the training code.

hence, the text the model generates is arbitrary. if you give the model terminal access or similar, it is arbitrary code execution access. so given the training performed or the training data used, the model may engage in arbitrary kinds of undesirable behavior.

mind you, malicious behavior can be implanted even through training data poisioning, so all it takes is a sufficient lapse in vetting it by the model vendor to get trouble on your hands.

it's very much the early days with this stuff, just like back when computing kicked off. the big deal was that "thing exists at all". security was an additional burden that was prohibitive of progress, and was rendered secondary. remains true here as well.

This was a very illuminating read that put into words something I've noticed but never quite pinned down. Thank you.

is this the two types of tech posts that will exist moving forward?

1. look at this (maybe) new thing i made by prompting an LLM that i now (may or may not be able to) explain, ghostwritten by an LLM

2. look at this thing i artisanally handcrafted because i'm such a badass, owning Big LLM and its subhuman followers so hard with this, and i totally do not feel persistently paranoid about being intellectually cucked by LLMs whatsoever

superoptimizers existed long before LLMs were even a naughty thought, and an LLM would have written one for this problem without any issues, likely without even any specific steering that way. which matters exactly none whatsoever if exploring superoptimization oneself was earnestly the goal.

it's hard to genuinely blame the author for (possibly) feeling compelled to frame things this way, so i deleted a previous version of this comment that was even meaner about this. but that only helps so much in offsetting how incessant this type of blogpost is. now if this framing was more of an intentional choice however...

I have considered this angle before, but landed on that it's not like anyone was ever actually going to do so.

I tend to think of the market kind of like a social media feed, i.e. engagement matters. Even if they're establishing these results at an overhead, they're doing so with money that otherwise would probably never be spent on these problems, or would only be over very long timeframes.

Whether this is sustainable is where what you describe comes in. And it may not be, but that doesn't mean it isn't impressive. Lots of cool technologies in history that failed to justify their stay.

Why the ellipses? It's not like they're sweeping this under the rug, it's pretty explicitly stated for which optimizations this applies. It's also a perfectly standard optimization strategy.

Their second example features it for example, but as an additional optimization that saves memory. The compute optimization part doesn't need this, and is what brings the little over an order of magnitude speedup.

It's not a trick. There's no such thing as "being right" overall, that only exists in formal logic (i.e. in theory): https://en.wikipedia.org/wiki/Solipsism

Right or wrong is all relative to a segmentation. You don't discover who's right or wrong through debate, you discover whether you can agree or not, and why. Formal logic sidesteps this by definining a common ground up front.

I'm committed to my position, and I do think it is "right". But that doesn't necessarily mean much. Same for you. If this was more commonly understood, I think that would indeed be nice. See also the "agree to disagree" you issued up top. This is what that actually means.

We got immortalization at home.

Yeah, happy to see. Wouldn't it be cool if the many computational hardness assumptions were no longer just assumptions? If the algorithms and implementations used would no longer require the holy blessings of arbitrary - if credentialed - researchers?

Or consider the algorithmic complexity classes underpinning all of this, not just P vs NP, but also whether quantum computers are actually a distinct class, since that is not proven yet either afaik. The reason we're moving to quantum resistant algos in the first place.

If these arguably very finite amount of monkeys allow them to pull off such campaigns so frequently and consistently, these monkeys kinda do got it going on.

Maybe that explanation covers the rest that wasn’t identified?

every comment mentioning "ai psychosis" is an amazing caricature. people whining about original thoughts repeating the same resident tropes over and over again. basically just throwing out insults. try that for sad.

same people usually who'd be livid about misusing mental health status labels, and torturing language in general. but it's okay, as it is for a noble purpose: insulting people whose opinion you currently feel sufficiently uppity and righteous about. that will raise the bar!

the "last bastions of human values" demonstrating the absolute bottom of the barrel human behavior and intellect. good bit worse than sad, really.

Yes, I do get that, I was simply taking an issue with the characterization, cause it's super catchy but conflates several distinct failure modes.

The model emitting nonsense is mostly the model just being terrible in general. The model answering slightly different each time is inference details, like temperature, seed, batching, hardware, etc. The model responding to the same inquiry very differently is the model being semantically unstable. These all exist but they are distinct, and social media isn't kind to nuance to begin with.

You gestured at the second previously, but what you describe now is mostly the former. I don't really experience this anymore, but I do typically use Opus, and work an area where it's clearly excelling in, so that's probably why. Every time I go back to Sonnet, the bullshit parade begins, no matter the amount of context provided, or the thinking effort level configured. Mostly tolerable, but I wouldn't leave it unattended, like you say.

The third I only recently got into testing, cause it's kind of rich to re-run the same task many times just to eval stochastic variation, and not even always viable. Tested Luna, Terra, and Sol at various thinking levels, multiple runs each on the same problem, perturbing it with a (genuinely) random bit of text injected each time. The famed stochastic parrot effect didn't really materialize, I got normal parrots. I wouldn't necessarily say this was a clear win in semantic stability though, cause the random bit of text wasn't semantically relevant. Maybe if I reworded the prompts proper each test, it'd have been a different story. But then that's not a clean test per se. Controlling for variables is hard enough even in normal research, try controlling for billions reacting who knows how, which is the whole issue to begin with, after all. So you can't properly test without better understanding, but you can't gain the understanding without proper testing. Catch 22.

You "repeat" here something for "convenience" that you never talked about in this thread in the first place ("Musk and Big Tech"), meaning you likely don't even know which thread you're in, or what a forum thread even is.

You hope not to have to explain how weather or nighttime changes the color of the sky, after I already mentioned the seasons sometimes painting it white, meaning you're unable to keep track of what was already said here.

You ask me if I'm "broken", and accuse me of inventing things about your life, clearly demonstrating that you fail to understand metaphors, examples, hypotheticals, and other forms of figurative speech. Or at least pretend to, only when it's me using these constructions.

You engage in goalpost moving with the Windows rewrite example, given the very well reported on examples for rewrites of countless old games and even entire suites of productivity tools, like Adobe software. You further accused me earlier of ad hominem, after engaging in ad hominem, then just made an appeal to your authority below, referencing a GitHub you do not link, not here, not in your bio.

You refer to physical things not being helped by a software, and somehow that being surprising to you. You explicitly make a claim that things are only awesome if they solve world hunger for children specifically. Then flip flop to fixing the climate, then flip flop to eliminating "chemicals" in food. This is an utterly insane worldview and there's nothing to disprove on it, for they are opinions, of the outlandish and nonsense kind, not arguments. You nevertheless do insist on wanting me to do so, of course, meaning the very basics of mathematical logic are somehow unfamiliar to you, despite your proclaimed programming experience (only logical statements have truth values).

You explicitly conflate advertising, evangelism, and cult behavior, clearly demonstrating your inability of grasping that as societies scale, information needs to travel, which makes advertisement, news reporting, promotional activities, and so on a necessity. You thereby fail to recognize you're laundering in your judgement regarding the nature of the promotional activity, even after many examples demonstrating so, and a logical derivation explaining how. People may feel compelled and motivated to convince someone of something, yet you explicitly only consider malicious or otherwise degenerate motivations as existing.

You fail to grasp basic dilemmas, like how an autopilot doesn't need to have zero death rate to be a potential human replacement, it just has to have better-than-human death rate. A corvid could make this judgement correctly, yet your historied programmer self apparently cannot.

And you ask me why I'd insult you? You wonder why your observations go unaddressed? The cause is one and the same, now robustly evidenced, and it's very simple: you're unable to process what I'm telling you. And even if you were able to, you'd still be unable to make a point coherently and cogently. Even knowing your epistemic limits escapes you.

I give up. I cannot reason with what doesn't.

Think about that for like two seconds.

Yes, randomness is involved, except you're destroying the snot out of it by biasing it to hell and back with those billion-trillion parameter models in front of the sampler.

Alternatively, check the difference between a typical ChatGPT prompt response, and the output of the following:

  import tiktoken, secrets
  
  enc = tiktoken.get_encoding("o200k_base")
  tokens = tuple(enc._mergeable_ranks.values())
  msg_len = 128 + secrets.randbelow(2048 - 128)
  token_ids = [secrets.choice(tokens) for _ in range(msg_len)]
  print(enc.decode(token_ids))

You'll note that they're not much alike. You may also note that with LLMs you can set the "temperature" parameter to zero, in which case the sampling is no longer randomized. So much so, that given various runtime conditions, passing in the byte exact input will return the byte exact output, every time. Possibly across machines, depending on the inference engine used.

Even if you mean it on a "higher level", imbued into the language emitted, do feel free to ask a model with disabled tool calling to give you some random numbers / words. You'll find that it isn't very random. Certainly wasn't when I last tried this for entertainment.

The same recovery paths and risk boundaries apply as always.

If anything, these "random number generators" (which they aren't, do not ever use them for that, obviously) are:

- even more cautious,

- more readily take, stash, test and use backups,

- check for and take recovery paths,

- do dry runs,

- consider interruptions,

- mine for more information,

- smoke-test more,

- and are better equipped to do both deeper and wider analyses when encountering problems,

than the typical person.

Reading this made me recognize that better interpretability yields not just better aligned but more compact models too. Which makes a lot of sense in hindsight, and even has some ironic ring to it when considering the unending amount of anthropocentric takes regarding what an "understanding" means.

What list? Am I missing a comment? None of your comments contained any lists, especially not ones with some remarkable parallels to "cult activities". What are you even talking about? You failed to say anything of substance so egregiously, I've been practically arguing with myself on your behalf!

> Let me ask you again, why are so many people making posts of any lkind about with extraordinary claims about how AI is wonderful ?

What is so difficult to grasp about the difference between evangelism and promotion?

The sky is whatever color, and you're set on being holed up in your room with the blinds down. How would you know if somebody didn't try to convince you to look outside and check?

Surely you understand the counterargument now? People just fucking like this shit, and want others to try it! That's all dude. It's really just that simple. No religion required.

People go, fork out $20 for Claude, try the fucking thing, and see that just like you see that the sky is blue, the thing is actually fucking awesome. So that gets them excited, and makes some of them even promote it, to tell others they should give it a try, because it worked for xyz, and might work for them too. This is cultish behavior how exactly? Is every social behavior a cult behavior to you?

Have you ever enjoyed a song in your life? A movie? A book? Never felt the desire to share it with others, and see if they like it too? Were these acts of religious evangelism too?

You seem confused. I can insult you and address your points at the same time just fine, and that's exactly what happened too. Why couldn't I?

> ad hom

The reason I listed off the tropes "your observation" is a rehash of, was to highlight how in itself it's simply an ad hominem.

If you needed that spelled out, here you go. I know, calling people religious [fanatics] being just an insult, tailor made for ad hominem? Now that'd be just way too low...

> Nobody needs to push the fact the sky is blue because this is a basic truth.

See also you a few lines later:

>> try in winter when the sky is white?

> Dont be silly. Not in Australia.

So is it a Truth™, or just a shared and durable experience? Because suffice to say, places other than Australia do exist, and so said [lack of an] experience does have limits.

Am I evangelizing the existence of a white sky? Or am I just defending an equally tacit fact, that the sky is not always blue, from being denied and lied about? Does this make me religious all of a sudden? I'd hope not.

> Here we go again, personal attacks.

No, the concept of an oracle, and the argument (sarcastically) laid out involving it, was not meant to be just a personal attack there. An oracle is a genuine philosophical instrument. Re-read it under that lens if you care.

Here's a different version if you remain distracted by the sarcasm within.

Suppose being right and wrong is an actual possibility, and that I'm wrong while you're right.

You try convincing me. Under your own logic, "truth doesn't need evangelizing", and all manners of "pushing" something is magically evangelism. Therefore, we have two options:

- by virtue of you attempting to convince me, I'll be rendered right, and you'll be rendered wrong

- the preposition is contradicted

The latter obviously doesn't compute, while the former invites semantic appraisal. And so doing exactly that, what I'm saying is that no, the act of convincing another person doesn't render the person doing the convincing automatically false, unlike what you're proposing. Your rule of thumb is unsound. Since you understood this as a caricature, I believe you agree?

The other, possibly more charitable way to read what you were proposing is a sort of majority rule, where a majority consensus means truth. I'd expect it to be self-evident that no, that is no logical guarantee either.

You're welcome to put forward an actual observation that isn't just a veiled ad hominem itself, preferably with an actual argument that can be materially replied to. Or you can admit to having been simply parroting a dehumanizing trope, without giving it even two entire seconds of thought.

You're proposing that if something is disputed, then that's for sure wrong. And that if something is being promoted, then that's for sure deficient. Is it really so difficult to see why that's absurd? Seriously?