Top stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Learning Programming in an Age of LLMs(ploeh.dk ↗)
    21comments
  2. Introducing System One Models and Jev(typesafe.ai ↗)
    407comments
  3. Mistral X Mozilla: Private, Multilingual AI Browsing(mistral.ai ↗)
    11comments
  4. Show HN: An e-ink frame that hears birds and draws them as 1800s illustrations(github.com/arnegiacomo ↗)
    209comments
  5. Apple Reference Image: A New Approach for Verified Photography(security.apple.com ↗)
    199comments
  6. Show HN: I made a flight simulator, except you're just a passenger(inflightsimulator.com ↗)
    116comments
  7. An update on Wayback Machine access(blog.archive.org ↗)
    274comments
  8. Doing Everyone Else's Job(yosefk.com ↗)
    40comments
  9. Gemini 3.8 Live and 3.8 Live Extended Thinking(blog.google ↗)
    273comments
  10. Building a Linux GPU Driver for the M4 Mac Mini in One Month(codyho.dev ↗)
    192comments
  11. Negativland, Culture Jamming, and the Art of Making Something New(blog.archive.org ↗)
    20comments
  12. EU chief opens door for Canada to become 'associate member'(bbc.com ↗)
    discuss
  13. German Rheinmetall open-sources its Battlesuite connected weapon system protcol(rheinmetall.github.io ↗)
    78comments
  14. Recreating Voodoo Graphics and a Late-1990s Gaming PC on an FPGA(nand2mario.github.io ↗)
    38comments
  15. A software thing I built: GPS on a 25MHz 486-SX(vcfed.org ↗)
    7comments
  16. Intelligence per Watt: Measuring Intelligence Efficiency of Local AI(arxiv.org ↗)
    5comments
  17. Why I'm still bearish on LLMs after Navier-Stokes(dank.systems ↗)
    271comments
  18. MartyPC – A Cycle-Accurate IBM PC/XT Emulator(github.com/dbalsom ↗)
    2comments
  19. Better routing, probe fixes, plugin updates in Freenet/Hyphanet 0.7.5 build 1507(hyphanet.org ↗)
    discuss
  20. We got admin access to Baseten's production GitHub(strix.ai ↗)
    159comments
  21. Datamimic – don't let your coding agent invent its own test world(github.com/rapiddweller ↗)
    7comments
  22. Saving Jet Fuel(marksblogg.com ↗)
    50comments
  23. Show HN: Capsule – Single-file web apps that save their data into SQLite(withcapsule.app ↗)
    141comments
  24. Tech Fascism Has Come for American Democracy(techwontsave.us ↗)
    2comments
  25. Let's make quality the norm again(forbrukerradet.no ↗)
    416comments
  26. Chopping up books when they're physically too big(mattkirkland.com ↗)
    170comments
  27. Learning to solve hard problems in RL for LLMs by never giving up(mnoukhov.github.io ↗)
    5comments
  28. Suspected sabotage causes major Netherlands rail disruption(bbc.com ↗)
    439comments
  29. WangNet – 1.8 MB, zero-dependency Numberwang adjudication in 11 languages(github.com/graafhenk ↗)
    58comments
  30. The Beauty of Roundabouts(gruhn.me ↗)
    131comments

Intelligence per Watt: Measuring Intelligence Efficiency of Local AI

47 pointsby 2d agoarxiv.org
5 comments
1h agoHN ↗

We propose intelligence per watt (IPW), task accuracy per unit of power

Stupid metric. It's not because a model is better performing that it necessarily requires more energy or compute.

1h agoHN ↗

I don't think that's what they are saying. In fact if they did the metric would be pointless. Rather they are saying by estimating that value on different architectures, one can find more efficient ones. They use open model to be able to remove unknowns. They aren't advocating for one model or another, only more efficient architectures.

1h agoHN ↗

We propose miles per hour (MPH), distance travelled per unit of time

Stupid metric. It‘s not because you spend more time that you travel farther.

s/

27m agoHN ↗

Unless I misread it, are they saying local GPUs use less energy?

That’s surprising, almost unbelievable, due to batching. Local is usually not batched.

7m agoHN ↗

Small models are much smaller than frontier models though, which is how they end up consuming less energy despite low batch count. (Though with local models growing strong agentic capabilities, batching becomes a reality with local models as well).