tux3

Ubuntu is a Debian distribution that releases more often. A faster update cycle for their packages is basically their entire reason for existing.

I may be biased as a Debian user, but if Ubuntu now reacts slower than Debian this may push some people to reconsider their choice.

You must have missed Mendeleev's less well-known Alphabetic Table of the Elements!

A promise like that just isn't binding, things can change in 2 years, the managers themselves may have moved and been replaced with different people. A bird in the hand is worth many birds three years from now.

"Proteins" is not one thing, it's a class of many different things made from amino acids. Same as the "peptide" craze you might have heard of, except proteins are bigger.

Antibodies don't have to bind only to proteins, it could be some lipid, carbohydrate, any molecule big enough. Most of the time it's some protein, but it's not an allergy proteins in general.

Advisories with the details are public:

https://github.com/php/php-src/security/advisories/GHSA-w8hw...

https://github.com/php/php-src/security/advisories/GHSA-6wmf...

Wikipedia patched it yesterday for the wiki itself, but most of their other tools are still down right now.

This leaked yesterday in part because so many people had advance notice.

Right, I think I'm with you that browser/no browser shouldn't be a factor. I think I'd like to hear from Rust and Go webpki people on whether we could help the ecosystem enforce CT.

My point is more that as a user of a high-level library like reqwest, I'd like to talk safer defaults when root CAs are in use

I hold eMudhra's private key. I generate a cert signed by their intermediate on my box. eMudhra didn't issue the cert, so it is not sent to any CT logs, and I have no SCT to put in the cert.

Am I missing what you mean by monitor?

I believe Android updates things like CA certs independently of their monthly security patch. I'm not sure Android 12 uses the same system, but on newer Android I think those are shipped via Google Play system updates, which don't require waiting for the phone vendor.

Lean tactics can handle obvious things automatically, for some value of "obvious".

I don't see how this happens in a C++ function, since breakpoint() is not noreturn you'd have to be doing something very strange for the next instruction to be outside the function. It's possible some debuggers will show the next line, and even then that seems like a debugger QoL issue that's fixable pretty mechanically. A soft breakpoint can come from CD 3 or CC, even if you don't get the instruction length from the CPU there aren't that many possibilities. I'm sure we don't need everyone to start inserting extra NOPs to get the desired behavior.

I don't think labs really mind Pangram, so outside of some niche open models that might want to try, I'm not sure anyone is maximizing for stealth. And open models don't have the advantage in a GAN war, since one is open and the other not.

Anthropic is barely doing the minimum to keep Claude's writing tics bearable to read.

Unfortunately the recording is missing audio for half the talk. A frequent issue for the plumbers livestreams this year.

https://www.youtube.com/live/OPRciCSsdS4?t=2031

Detection tools in general have a bad selectivity and sensitivity, but there's consensus that Pangram is generally accurate and reliable. I don't use these tools much, but that's the word among people I trust, and that's what I see when I try it.

And, well, neither of us really knows the ground truth. Maybe the guy who has a repo filled with articles co-authored by Claude is actually not posting slop this time and HN users are wrong and Pangram flagging it is just a false positive. But I can't blame anyone for being tired of giving this the benefit of the doubt.

I'm ESL. The mistakes I make are part of what makes me human. If HN'ers are having an allergic reaction at the first hint of slop, it's because we've been drowning in it for months.

The recently released models have improved their writing styles. I don't know about this article, but if that's any indication of their writing habits, the author's github pages repo is full of posts generated by Claude.

A good first input in general probably looks like a bad password, like 123abcXYZ! with a little bit of every type of character (lower, upper, num, symbols)

(The NIST SP 800-63 Standard Password "Fall2026!" also works, but it's less efficient, and changes every quarter!)

There's many fields where it's easier to find new problems than to solve them, and it's easier to check the solution once you have it.

They're doing RL on open problems these days, not just next token prediction.

Correct! It's still an inode, even without a path.

Followup quizz:

A file is held open read only by a process, and then you delete the file. Under memory pressure and without swap, can Linux evict those pages to reclaim memory?

The Default trait provides a function that actually constructs the type in question. But here the ! type can never be constructed, so the only way to implement Default would be to have it panic, loop infinitely, or otherwise fail at runtime.

So this would risk turning a compile-time error into a runtime error.

Of course. Auto-vectorization isn't magic, for one. In any video codec, the hot kernels will all be in hand-written assembly.

Now I'm not saying AI can't write ASM. I'm saying that compilers have a hard job. They output bad assembly, because the task is difficult.

They're saying the weekly numbers grow a little bit every week, so every week is technically record breaking.

The absolute number of lifetime downloads can of course only go up (but that's even more vacuous).

As the saying goes, the optimal amount of cheating is not zero.

I'm saying it's variable, not that cheating can disappear completely.

In other words, utopia is fictional, but that doesn't mean the real world couldn't be better.

I see, my bad. I tried to follow the comment and jumped to conclusions, sorry about that.

I admit I'm not a fan of the loop macros, or writing constants like 0xd80000 in decimal, or having all of that code inlined with no further explanation or git history.

But I think it's just that this file is lifting parts of the logic from PAQ8 (the inlined num_states/next_state named in a comment sure look exactly like PAQ functions translated into C, which, kudos to PAQ for having extensive comments in their code by the way).

I'm a bit disappointed that PAQ isn't mentioned in the README, and only very obliquely in the code, if that's where those functions come from. PAQ is GPLv3, at least a mention could have been nice (assuming I'm not wrong about the source of that logic).

If OP is willing to cooperate, I'd like to offer an independent investigation on site, and I will be bringing pizza

So, it looks like this tries to build a model (or rather, multiple layers of models combined with something called context mixing[0]), potentially analyzing the whole file first? Is it fair to call this a very fancy entropy coder, replacing the last phase of what a traditional audio codec does (the lossless part, after quantization)?

I like the code style, this is pretty nice C. It's refreshing to see human code.

[0]: https://en.wikipedia.org/wiki/Context_mixing

Short statement from the WMF in response: https://wikimediafoundation.org/news/2026/09/03/us-staff-uni...

They seem to say they'll accept the outcome and pledge to engage in good-faith. Which is the only correct response, really, so time will tell if these words are more than generic PR.