b10430
llama : allow virtual igpu devices (#26953) llama : allow virtual igpu devices cont : better comment Website: https://llama.app macOS/iOS: macOS Apple Silicon (arm64) macOS Apple…
Full feed
Feed 27,292 posts
Every story across every category, newest first. Each card links to the original publisher; daily-brief posts open as editorial pages.
HashAgent – Share an AI agent as a URL, runs locally via WebGPU
Article URL: https://hashagent.pages.dev/ Comments URL: https://news.ycombinator.com/item?id=49298088 Points: 5 # Comments: 0
fantastic: latest llama.cpp server webui can now run commands for tools into rootless sandboxed containers
while having much fun testing LLMs Houdini-like attitudes and abilities to evade, excalate and escape from carefully reciprocally arranged security enhancing sandboxing VMs, containers, namespaces and…
Show HN: We Implemented the IPv8 Internet-Draft in Linux, Libc, and BGP
Article URL: https://goonhost.rocks/blog/implementing-ipv8-internet-draft Comments URL: https://news.ycombinator.com/item?id=49297996 Points: 12 # Comments: 2
Are we getting Qwen 3.8 35-A3B?
So far, it seems like Qwen 3.8 might drop today as a 27B dense model. If that’s the case, no MoE offloading this time , I…
Dear people who work at the airport
Article URL: https://life-after-ssri.bearblog.dev/dear-people-who-work-at-the-airport/ Comments URL: https://news.ycombinator.com/item?id=49297801 Points: 81 # Comments: 50
b10429
server: allow accessing /metrics and /slots during llama_decode() (#27041) server_queue::worker call llama_decode inside yield_to_queue also handle process_mtmd_chunk clean up nits rm test Website: https://llama.app macOS/iOS: macOS…
What KV cache tricks do you guys do with your agents?
(Typing on my phone, apologies in advance for my “shorthand”.) I recently learned about prompt caching in llama.cpp. Basically, it’s a setting where your kv cache…
Frontier agents don't comply with standards, even when instructed to
TLDR: Our open testbed LARA examines the behavior of frontier LLMs in realistic agentic deployment contexts. Previous results showed all models routinely take actions that would…
Specification-first convergence with an AI coding agent: a case study of dismantling a core architectural invariant across 189 files in a 717k-line codebase with no test oracle and no human code review
This paper reports a single, fully instrumented case study of a large-scale architectural refactoring by an AI coding agent under a specification-first protocol, with no human…
The TEMU-Fication of Software, Digital Goods and Services
Article URL: https://xn--gckvb8fzb.com/the-temu-fication-of-software-digital-goods-services/ Comments URL: https://news.ycombinator.com/item?id=49297637 Points: 8 # Comments: 0
Cursor is joining SpaceX
Cursor has officially been acquired by SpaceX.
Claude Code now runs daily maintenance on Anthropic's software with a 46 percent merge rate
Anthropic is testing whether Claude Code can handle daily maintenance of the company's own apps, from crash fuzzing to dead-code removal. In a few weeks, the…
WhatCable: Know what your USB-C cable can do
Article URL: https://www.whatcable.uk/ Comments URL: https://news.ycombinator.com/item?id=49297469 Points: 6 # Comments: 0
A modest community proposal for desloppification
Over & over I see people post slop and excusing it as "english is not my native language". As if, if we could understand their language…
LFM 2.5 2.6B is the best small model for tool use I have ever used.
I'm working on a local perplexity/AI search comprised of a custom harness and a further trained model. LFM 2.6 is almost to spec with no additional…
Don’t forget why learning is important
This is a timed post. Every 5 minutes while writing this post, I had to stop to do 13 push-ups, and when I could no longer…
Everyone talks about AI agents. This is what one looks from the inside
Article URL: https://pssah4.github.io/vault-operator/guides/capabilities Comments URL: https://news.ycombinator.com/item?id=49296964 Points: 3 # Comments: 0
For the love of god stop using CPU limits in Kubernetes
Article URL: https://github.com/inevolin/k8s-cpu-limits-analyzed Comments URL: https://news.ycombinator.com/item?id=49296939 Points: 11 # Comments: 0
What If We Enforced AI Model Safety At the Level Of GPUs?
Tldr:AI Agents (e.g. based on models like Claude Opus and Fable) are now powerful enough to be used as autonomous tools for large-scale cyberattacks. This most…
Qwen 3.8 Max Is Extremely Good at Knowing What Not to Build
Qwen 3.8 Max is out, and after spending time with both the early API preview and the full release, I’m pretty impressed. The full release mostly…
Ntfy – open-source Push to Mobile
Article URL: https://ntfy.sh/ Comments URL: https://news.ycombinator.com/item?id=49296902 Points: 19 # Comments: 5
Zhipu AI releases GLM-5.3, claims it's the strongest open-weights coding model
Zhipu AI has released GLM-5.3, a model that, according to its own benchmarks, is the most powerful open-weights coding model, with a 50 percent improvement over…
Getting ready for the big 3.8 drop (strix halo centric but applies widely)
https://sleepingrobots.com/dreams/mtp-qwen36-strix-halo/ This is slightly out of date but the eli5 is that MTP is awesome on these qwen models. Since 3.8 is said to be the…