Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Show HN: Skillmem – local memory for coding agents that stores how, not what(github.com/liza-studio ↗)
    discuss
  2. AI is an elite crime spree(thebignewsletter.com ↗)
    discuss
  3. Fibonacci Product(johndcook.com ↗)
    discuss
  4. Ask HN: What happens when Claude Code goes away?
    discuss
  5. With a Shovel of Salt: AI Success Stories and the Retraction Gap(misalignedmag.com ↗)
    1comments
  6. Show HN: pg_raw_parse, parse Postgres SQL in Rust 20-60x faster than pg_query.rs(github.com/pgdogdev ↗)
    discuss
  7. Key Alzheimer's study cancelled amid Trump attacks on science(theguardian.com ↗)
    discuss
  8. Human brain is two separate organs, research finds(medicalxpress.com ↗)
    2comments
  9. Textured(pluralistic.net ↗)
    discuss
  10. Gov. Newsom issues executive order to rein in AI 'before it's too late'(cnbc.com ↗)
    discuss
  11. Cutting our agent's harness cost 10x without breaking the product(usenotch.ai ↗)
    discuss
  12. OpenAI 'ethically hacked' with help of Anthropic's Claude chatbot(theguardian.com ↗)
    discuss
  13. GrassLobster: AI Agentic Generation of Parametric Geometry Workflows(miro.vision ↗)
    discuss
  14. You could have built Jev(sgnt.ai ↗)
    discuss
  15. Coffee and Milk ≠ Latte(johndcook.com ↗)
    discuss
  16. A custom virtual machine for the Stars 4X game(nullprogram.com ↗)
    discuss
  17. What happened on Voat after Reddit banned major communities(nature.com ↗)
    discuss
  18. Build Faster Feedback Loops Using Qualitative User Research(nseldeib.com ↗)
    discuss
  19. Show HN: Omabib, a Zotero alternative for Omarchy with agent support(github.com/atomashevic ↗)
    discuss
  20. Are You Ready for Part 108?(marbletech.us ↗)
    discuss
  21. Show HN: Rickub – The Smartest Git in the Universe(rickub.com ↗)
    discuss
  22. Why Harvard's Dean 'Encourages' Students to Use AI(derekthompson.org ↗)
    discuss
  23. ThinkNode M9 MicroSD Card and Virus Notice(elecrow.com ↗)
    discuss
  24. Let GPT-6 Astra code without using Codex Usage(ethanplus.ai ↗)
    discuss
  25. Show HN: Quokka – a self-hosting, deterministic programming language(quokka.space ↗)
    discuss
  26. Show HN: Servers for AI, not another AWS(rawhq.io ↗)
    discuss
  27. Converting between cosine similarity and concentration ratio(johndcook.com ↗)
    discuss
  28. How to Secure Coding Agents(augmentedswe.com ↗)
    discuss
  29. 'My Husband Has No Friends'(thecut.com ↗)
    1comments
  30. AI Protest in Montreal(montrealgazette.com ↗)
    1comments

Neso: Metal MSL and D3D12 HLSL Back End for Triton

1 pointsby 1h agogithub.com
1 comments
1h agoHN ↗

Neso is an ahead-of-time compiler backend for Triton. It lowers Triton TTIR/TTGIR into MSL or HLSL shader code, to allow modern LLM compute kernels such as Flash Attention II to run on GPUs in consumer devices such as Macs and Windows laptops. A companion project, NSpeech https://github.com/dropbox/nspeech, built in Rust with Candle, demonstrates how Neso can be used to power local ASR and TTS models.