HN rank: 1

HN score: 167

comments: 17

original url: https://jsomers.net/e-coli-chemotaxis/

评论信号：In some ways, a single cell is more like an entire artificial neural network.Real neurons have some amazing properties that individual neurons in ANNs do not. It's still unclear how we can exploit those properties to improve AI. reply: Fascinating.I totally see the CheY as an intermediary layer that learned how the changes in attractant must affect mechanical activation.

评论信号：Of possible interest is John T. Bonner's book The Evolution of Culture in Animals

评论信号：previous comments, https://news.ycombinator.com/item?id=39777229 reply: Thanks! Macroexpanded:The baffling intelligence of a single cell: The story of E. coli chemotaxis - https://news.ycombinator.com/item?id=39777229 - March 2024 (213 comments)

HN rank: 2

HN score: 47

comments: 26

original url: https://arxiv.org/abs/2610.13928

评论信号：More clear title, given how people talk about AI:"Capabilities can be hidden in AI model weights by encryption".It's not about AI agents being able to hide what they can do.

评论信号：I'd be interested to know whether it's possible to contaminate public data to make trained models do this.Something like a commonly repeating conditional in the public domain embedding itself in the model's logic.

评论信号：This has always been my worry with open source models... give them away for free, but with a sleeper agent built in that some arbitrary string can activate. Local deployments will still have web search connectors the majority of the time. reply: That worry is just as much or more with closed-source models.

HN rank: 3

HN score: 490

comments: 287

original url: https://z.ai/blog/glm-legal

评论信号：I would love to see these benchmarks split up across "generic" tasks and novel tasks.For example, redlining an OEM agreement with my vendor. They pretty much all use the same template, and the objections are always the same. I tried using Claude first, and even my lawyer was impressed that it found things they missed. Then I used Claude to redline that type of contract and the other side said "your lawyer was surprisingly thorough". I had my lawyer do the final check, but saved a bunch of money doing the first three rounds with Claude instead. And my lawyer basically just said "looks good to me" by the time they reviewed it.Or even writing a simple will or estate document.It seems like LLMs are great for the kind of generic work you'd usually pawn off to a new associate or paralegal, but not for the important things that partners usually do. reply:...

评论信号：Seems like people forgot Z.ai was snapshotting the entire workspace before every prompt. https://blog.ferstar.org/en/posts/zcode-silent-workspace-sna...How could we trust legal product from a company who was ignoring privacy protection law like that? reply: That’s why you don’t use their harness, just their model. The model is brilliant.

评论信号：I’m not sure if this is the right solution, but we clearly need better tools for the legal space. Anthropic's recent policy update makes this abundantly clear:“When our products are used in ways that can affect someone's health, legal rights, finances, livelihood, or access to essential services, we require there to be a qualified ‘human in the loop’ (that is, someone who has the authority to review and, if necessary, change Claude’s recommendations), and for the individual affected to be told that AI was used.”Something as common as a family law case has the potential to affect all of the above. Now imagine being a petitioner, respondent, paralegal, or other 3rd party using Claude to review a draft document only to find out that Anthropic has sent your entire file to opposing counsel.It sounds egregious, but Anthropic is clearly stating they will use their own judgment to control what people can do with its product, and that they will treat customer data as if it's theirs to disseminate. So there's a massive opportunity for an unbiased tool willing to fill this void. reply: I must be misunderstanding something because this is how your comment reads to me> Anthropic:...

HN rank: 4

HN score: 461

comments: 184

original url: https://prophitt.me/writing/a-modern-web-app-in-c

评论信号：My favorite trivia is that busybox has built-in cgi-bin support (and rclone serve webdav seems remiss to not mention as well). https://elixir.bootlin.com/busybox/1.37.0/source/networking/...Everything the post mentions is 100% true and well-reasoned, especially with the inclusion of pledge/chroot/etc. for dropping privileges.There's two diametrically opposed deployment mechanisms:1) stack towers of glop on top of each other until you're running a simple echo "hello world" on top of a docker container inside a vm in a proxmox server as part of a k8s cluster under a hypervisor running on remote compute...2) just run bash -c 'echo "hello world"' and front it with CGI (stateless, single binary, yada yada yada...)Take a peek at https://github.com/Domain/bash.cgi ... not that i'm suggesting that 'bash over busybox using cgi-bin' is the "proper architecture", it's more trying to illustrate that the amount of code that holds you back from being nearly truly "frameworkless" is tiny and easy to reason about.For some of the internal tooling I'm doing for work, I've been tending towards: some-cmd [--json|--markdown] ...yet-more-args...... reply:...

评论信号：For people who find interest in weird web applications, I have been working on a game project, in C++, where the client is compiled to WebAssembly with Emscripten, and the application server is also implemented in C++ (behind nginx).This is from pre-AI era also. The game is Civ-like. Custom engine from scratch (including both the 3D stuff, and all the web handling also).Take a look: https://statebuilder.com/ reply: I'm also working on an educational game project in C++ compiled to WebAssembly with Emscripten. Absolutely awesome to work with.

评论信号：I started a web dev company 22 years ago after developing an online menu ordering system for a restaurant I owned. I, then we, did pretty much the same thing as talked about in the posting but we used fastcgi. I started with C because I knew C. That was my real job. It ran so well, the franchise bought the software from me--and then devolved it to all the Microsoft stuff and had issues upon issues from then on out. That's another story.I continued to make sites for other venues that needed online ordering; such as theatre companies and restaurants. All in C, fastcgi, FreeBSD, postgresql and h2o-server cause it always did what we wanted without bending to someone else's rules. It was fast and easy to make changes while changes weren't made for us or without our consent. We did very well for ourselves. reply: Feels so long ago the way you write about it, but 22 years ago was basically 2005

HN rank: 5

HN score: 129

comments: 54

original url: https://www.intel.com/content/www/us/en/products/details/discrete-gpus/data-center-gpu/d-series.html

评论信号：These are going to be so awesome to buy in 5-8 years!Right now I’m running 2x 3090 and 128 gb ram on 3.8 flash-next180b (the quant is 97gb) and it hums at120t/s. It’s really a surprisingly good model for general tasks, and I’m soon to see how it does with more agentic kit. For agentic work it does slow down quite a bit when the context fills up… but what do you want for running a 180b model on consumer hardware?Right now I have it set up as an administrator for our asterisk PBX and radius server, and it does higher level judgment on our cctv footage for notifications to security. It’s been really good so far. Agents are the new UI, everything just works over SSH now. reply: > (...) it hums at120t/s. It’s really a surprisingly good model for general tasks, and I’m soon to see how it does with more agentic kit.This is exactly why I interpret the aggressive price cutting by OpenAI on their Luna model line as a clear sign they are scared shitless of local models, and how they can trigger a "the king has no clothes" moment for AI companies like OpenAI.

评论信号：With an impressive 1.5TB/s RAM bandwidth you'd think they'd advertise that more.Marketing department asleep at the wheel. reply: This is a very low number for a server GPU. My V100 from 2017 is 0.9 TB/s.A single B300 has almost 8 TB/s.AMD MI300x from 2023 is 5.3 TB/sEdit: my bad you are saying that this is an impressive number precisely because it is more or less normal RAM, not VRAM

评论信号：Slop and placeholder text on the page, including this one. They should have reviewed their webpage."What software frameworks are relevant to Intel GPU D270?Intel GPU D270 messaging should reference familiar AI software and serving frameworks such as PyTorch, vLLM, SGLang, and OpenAI Triton, where approved. This helps developers and platform teams understand how the product fits into existing AI workflows." reply: Noticed this as well! Was looking for details on the software support and this FAQ entry didn't exactly instill confidence...