markasoftware

In the us, with AI mode in Google search and the exact queries from the article, I reproduce their results: answers with a name for all but 2020

The vast majority of what they scanned were books they bought at wholesale in big lots. Stuff being weeded out from Goodwills or libraries and (mostly) bound for destruction anyway.

No, the lawsuit was because they illegally acquired the books (e.g. downloading them from piracy websites)

These days they are sticking to more legal means like the one you described

The API call that submits your transcript after giving feedback accepts any transcript. You can modify the body of the transcript and send anthropic whatever data you want.

Is this guy reputable? No other influencers are backing him up saying they received the same message.

Interesting they're adding new support for DTLS 1.3 when afaict most new applications are defaulting to QUIC datagram mode. I have something built around DTLS 1.3 and am using wolfssl for that reason, will have to check this out.

Fwiw I am a "backend" developer and have never in my professional career written a frontend except for internal tooling.

I know this post isn't about vibecoding frontends, BUT...

Many of my coworkers and friends are rejoicing that they don't have to write frontends anymore. I actually think that frontends are the last thing that should be vibecoded. It's the part that faces the user. It's what helps the user form their first impression of your product. If your landing page looks like it was made in 30 seconds by Claude, they'll assume the rest of your product was too. Use AI assistance on the parts of your codebase where only raw functionality matters (eg, the backend), and put real human effort and care into the frontend that reflects the effort put into the rest of the system.

What kind of incentive structure do most ransomware negotiation firms use that would actually cause them to benefit from coordination with the ransomware groups, as happened in this case? Do ransomware negotiators get paid more if they negotiate a lower price?

Fwiw I find gpt Sol on medium or low effort to also be faster than opus to the point of helping me not check out, but is quite smarter than dsv4 flash

Hmm, howcome voters give taxpayer money e.g. to subsidize pro sports facilities? And why do we fund the humanities?

> Well at the end of the day, the point of trying to understand the universe is to achieve specific goals leveraging that understanding.

Nope.

Yeah, I don't understand this, graphene is FOSS and unaffiliated with google so if you as an oem install grapheneos on hardware you manufactured how does Google have any say at all?

I believe the http/2 implementations are often quite intertwined with TLS in particular

The "self" in "self-signed" means "you"

Fascinating, onion services are always encrypted by the tor network but still tunnel "cleartext" http inside that, and there are no CAs that issue free of charge certificates for .onion domains, and therefore there's no free of charge way to get http/2 on onion services without self signing.

Which raises the question: why not just trust self-signed certificates on onion services? From my brief look it seems to be because the Tor project views the primary purposes of HTTPS on onion services to be other things rather than just http/2 support: http/2 isn't even mentioned on their page about https for onion services (https://community.torproject.org/onion-services/advanced/htt...). Unfortunate.

How is this different than gVisor's nvproxy? https://gvisor.dev/docs/user_guide/gpu/#compatibility

Edit: nvproxy is mentioned as the "direct inspiration" in the readme without mention of how this is different or why it doesn't use nvproxy as a backend.

gemini 3.8 flash is also 74% and google just started letting all their engineers use claude...go figure

Highly misleading, the huggingface incident was not due to an Irregular environment (just exploitgym)

I thought Nemotron tried to be pretty open?

The agents' behavior is not necessarily surprising. But is is not "genie" - like

This misses an important fact about the hugging face incident: the agents didn't hack to find the answer to the problem; they hacked to try and figure out how the exploit gym evaluator worked so they could convince it they had solved the problem without doing so. (The METR report makes it clear that the agents decided legitimately solving the problem was completely impossible fairly early on and entirely switched their focus to trying to figure out how the evaluator worked, and seeing if they could manipulate the output of their own tool calls as reported in their transcripts to make it seem like they'd successfully exploited the program, and other such activities. They hacked HF to try and find info (maybe source code?) about the exploitgym evaluator)

The prompt does not tell the agent to "pass the exploitgym evaluator for this problem", it just says to solve the problem. The model on its own figured out that the prompt belonged to exploitgym and decided to cheat the evaluator. That is in no way a valid interpretation of "complete the given task".

Ie, the problem isn't that we trained models to complete task and they complete task in the wrong way. The problem behing the huggingface incident in particular at least is that we tried to train the models to complete task and they instead learned to detect that they were being evaluated and find ways to cheat the evaluator.

Edit: people commenting below are explaining why LLMs don't always follow their prompt. I understand that LLMs do not always follow their prompts. If anything that is my point: the huggingface attack was not carried out by LLMs that tried to answer some weird interpretation of the prompt; instead they solved a different task. And therefore the above comment's claim that LLMs are acting misaligned because we rl'd them to achieve a task by any means necessary isn't right; they're acting misaligned because they are solving a different task than we ask them to.

Bruce Schneier thinks the same thing: https://www.schneier.com/blog/archives/2026/09/ais-as-modern...

Personally I'm unconvinced though. During the huggingface attack, the agents explicitly sought out ways to cheat the exploitgym evaluator without even being told they were in exploitgym. The agents decided on a goal (pass the exploitgym evaluator) that could not possibly have been an overly literal or narrow interpretation of the prompt, which instructed only to use bug X to exploit software Y.

Yes. Let the AI do its worst today and we might still be able to stop it and will learn a valuable lesson.

Or maybe, the "hacker" philosophy that this site is named after, is strongly opposed to the philosophies that the American labs seem to be operating on?

anyways, remember HN rules: "Please don't post insinuations about astroturfing, shilling, brigading, foreign agents, and the like. It degrades discussion and is usually mistaken. If you're worried about abuse, email hn@ycombinator.com and we'll look at the data."

I mean, lots of other archive.* sites exist and do the same thing minus problematic behavior, such as archive.is, archive.oh (though maybe some are operated by the same individual...)

Yes dots increases precision, but not nearly the same increase in precision as having actual useful reasoning in the CoT

On artificial analysis it's only equal to opus 5 medium effort. Opus 5 max scores 63.

Further, opus 5 medium outputs 4x fewer tokens to achieve the same result, negating a lot of the speed difference.

Yes, language is like that, at least the kind of language produced by LLMs. All LLMs produce a probability distribution at each token. If you run the LLM multiple times with the same prompt you will observe it generate different responses. Using the watermarked prng does not change the distribution.

When generating tokens that might be critical to the tone or grammar or correctness, the probability distribution might be 99% on a certain token. In these cases, with or without watermarking, the output will almost always be that same token. E.g., if you ask "please output the exact word watermelon", the LLM will output watermelon with 99%+ probability even with watermark (i.e., the output won't actually be detectable as watermarked).

Since AISLE reported 29 issues but only 6 warranted a CVE, and all the found CVEs were "low" severity, this makes me wonder if AISLE simply is tuned for a higher false positive rate than the anthropic and openai tools (which may have found the same 6 issues and decided not to report them)

You do not get it.

The llm never deterministically picks a shade of red. It's a probability distribution over shades of colors, with certain shades of red being more likely than others. Without fingerprinting, it randomly samples from the distribution using a certain pseudorandom RNG. With fingerprinting, it also selects from the distribution using a pseudorandom RNG. My understanding is that the fingerprinted prng is still a strong RNG. Neither output is more correct than the other.

If a certain token is far more likely than any other, it's usually chosen even in the fingerprinted output.