HN rank: 1

HN score: 446

comments: 265

original url: https://arcprize.org/results/deepseek-v4-flash-0731

评论信号：I've been using it extensively since the release and the best summary I can give is that it's good enough to use it for (almost) everything and cheap enough that the cost are irrelevant. I'm running it in Oh My Pi with a second instance running as "advisor" and even with 5-6 active sessions (effectively 12 streams) I'm struggling to spend more than 5 bucks per day.OpenCode Go even has double limits temporarily so for 10 USD you effectively get 140 USD of tokens to spend. It would impress me if someone could burn that amount with "normal" usage. Even when running multiple sessions.I have a Claude Max subscription but I've barely touched it, it just feels like a step back to have to think about limits and usage even though the models are stronger.The beauty of intelligence at this cost (even if it's not SOTA) is that it opens a whole bunch of new use cases. Test failure in CI? Have the bot automatically propose a fix, its cheap enough that you can discard it w/h issues. Test coverage too low? Auto generate tests on CI for every pull-requests!... reply: If what you're saying is true and accurate, then US-based AI labs are in big trouble....

评论信号：Note this is the 07/31 release of DSv4 flash and not the "preview" that they put out a couple months or so ago.I've been running this model locally for a week, and the preview version before that. This updated one feels like a whole tier up. It's very capable for debugging and analyzing documents/data I upload.The killer feature, IMO, is the speed. On 2x RTX Pro 6000 Blackwell, its ~8k tok/s prefill and ~250 tok/s on a single stream. I saw 1000 tok/s with ~64 concurrent streams on vLLM.That's fast enough that you can interactively chat with it without switching tabs while you wait, and its a ~300B (13B active, hence the speed) model so the responses are also very good. It's actually more convenient now for me to direct 95%+ of my day to day usage to my local model, and only use Claude Fable for really big coding tasks.Until this model was released, I was contemplating spending even more money on hardware to run GLM5.2 (~750B) at reasonable speeds, but I no longer feel that need. This is smart enough, and I think it only gets much better for local models from here. reply: What quantization level is that? Because official endpoints are slow.

评论信号：My Claude account was banned the other day. The only possible cause I can think of is that I tried to authenticate from the AI assistant in a JetBrains IDE and, not thinking, entered the details for my regular subscription rather than an API account. As soon as it became apparent that I needed an API account rather than a subscription, I just closed out of the tab. Nevertheless, about 20 minutes later I got an email saying my account was banned for a violation of the usage policy, and my appeal was rejected.My initial thought was to sign up for ChatGPT, but I had $20 in OpenRouter so I've been trying out DeepSeek V4 Pro with Pi for the last few days and I gotta say, it's good enough for my use case. And even with paying for API usage rather than Claude's subsidised subscription, and with OpenRouter taking their cut, I will probably end up paying significantly less overall.... reply: I'm the same way, I have a very low/sporadic usage of any subscription I've tried. I now just use openrouter with DS4 pro/flash. It also gets rid of usage anxiety where I would try to justify the $20/month by forcing myself to use the tokens for projects as the weekly limit deadline neared.

HN rank: 2

HN score: 65

comments: 25

original url: https://genesisopenmodels.anl.gov/

评论信号：Just realized that there are basically no American open models right now ever since the Llama series was abandoned. Basically Gemma and GPT-OSS I guess?Ah but Mira Murati's new Inkling is Apache 2.0But it makes sense that if you're a university researcher you are thinking about what's a model that will be open weight and developed over the long term and doesn't raise 'Chyna' concerns in Washington DC reply: There's a bunch of American open models. Inkling, Nemotron, Trinity come to mind, but I'm sure there's others.

评论信号：Does Europe have an equivalent program?

评论信号：What would the selected participants get from this? Looks like there is no offer of funding?

HN rank: 3

HN score: 21

comments: 21

original url: https://www.economist.com/science-and-technology/2026/08/06/should-ai-labs-be-treated-like-the-owners-of-dangerous-animals

评论信号："The law in America relies on intentionality, notes Rune Kvist, head of Artificial Intelligence Underwriting Company, which insures AI firms. If no human intended to hack anyone, no crime can have happened. The ability to sue for damages is limited too."This is sloppy. Criminal negligence exists (although some crimes do require intent). And civil tort certainly doesn't require intent.

评论信号：I don't get it. Why is breaking the law so hard to enforce when it is a company (i.e a person or group of people consenting to) running a computer program? If I take a gun and spray bullets around me I don't get to write it off as the gun being dangerous. reply: But if you told an embodied AI to do a home cleaning task, and it decided to pick up a gun and start spraying bullets, you might not want to be held liable for that.

评论信号：> The hacks also present a challenge for legal systems. Hacking, when humans do it, is a crime. When an ai is the wrongdoer, though, it is unclear how to assign blame.Blame the prompter or person who assigned the task to the AI. It's their responsibility to use the tool in a safe way, just like it's a gun owner's duty not to fire their weapon carelessly into the air. reply: How are you going to figure that out? You can’t capture an agent in a jar and convince it to confess. As we saw with the HF incident, even highly sophisticated actors need a good chunk of time and manpower to trace these things. And I suspect the folks who are going to have the most success with LLM-powered cybercrime are going to know how to cover their tracks reasonably well.

HN rank: 4

HN score: 243

comments: 55

original url: https://github.com/xoreaxeaxeax/asm-hall-of-shame

评论信号：Related, and linked in the readme: https://github.com/xoreaxeaxeax/smiiiiiiiiiiiiiiii (using the slow instructions to break SMI) reply: I wish they would just explain it in normal terms instead of this nasty LLM "engaging blog post" style

评论信号：It says in the rules> Trapped/emulated/virtualized instructions may only time the trap, not the handler.But I feel like that 12ms write to an ACPI IO port at current leaderboard position 8 is probably trapping to SMM and being handled there.

评论信号：Nop should be #1, because it is infinitely slow for what it does. ;) reply: Strategy: nop does nothing. It opens the leaderboard accordingly.Score: 1 cycles Time: 0 nanoseconds

HN rank: 5

HN score: 142

comments: 47

original url: https://ancientlibrary.net/

评论信号：Interesting project. I would love if you could switch fonts to something like New Athena Unicode.I built something similar to this by cloning the Diogenes repo and getting Claude to re-implement it in Python (it’s a very old battle tested Perl code base, so a great reference implementation) and using the TLG database for the Greek and Latin texts. You can take this even further by integrating it with the Barrington Atlas (there are scans on Anna’s Archive) for looking up ancient place names, so you have dictionary + map lookups. If you’ve read any of the Landmark series books you’d known what I mean.Also better if you can generate chapter by chapter critical apparatus on difficult grammar and Anki decks. It’s an annoying part of learning these languages to have to stop and look up words, I like spending a few days learning vocab before reading and it makes it so much more pleasant than having to stop and look things up the whole time.Obviously all this stuff is copyright so it can never be shared, but I don’t care it’s for my own personal use.... reply: > LLMs have been an absolute game changer for me who is a hobby classicist but also knows how to build software.Definitely!...

评论信号：I'm always surprised to see that a large enough portion of the HN community are interested in classics that posts like this make it to the front page. Who are you all? What are you doing here?My undergrad was in classics and viticulture, then did grad school in atmospheric science, which led me to tech and hackernews.How did you get here? reply: I'm a recently retired software nerd who got into reading 19th century literature. I was greatly inspired by characters in these novels, most of whom had the advantages of a classical education, and I lamented how poor my own was. One day I idly wondered how hard it would be to learn a bit of Latin (as I was constantly having to look up Latin references) and once I discovered LLPSI[0] I was hooked.[0]: https://en.wikipedia.org/wiki/Lingua_Latina_per_se_illustrat...

评论信号：Neat! Similar to my graduate thesis, NoDictionaries: https://nodictionaries.com/cato/de-agri-cultura/156 reply: Nice.May I point out that in "Crudam si edes, in acetum intinguito", that "edes" is more likely to be the future of edere/esse "to eat"? (Just guessing by context.)