Top stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Introducing System One Models and Jev(typesafe.ai ↗)
    388comments
  2. Apple Reference Image: A New Approach for Verified Photography(security.apple.com ↗)
    149comments
  3. Show HN: An e-ink frame that hears birds and draws them as 1800s illustrations(github.com/arnegiacomo ↗)
    203comments
  4. Show HN: I made a flight simulator, except you're just a passenger(inflightsimulator.com ↗)
    79comments
  5. An update on Wayback Machine access(blog.archive.org ↗)
    265comments
  6. Doing Everyone Else's Job(yosefk.com ↗)
    25comments
  7. Gemini 3.8 Live and 3.8 Live Extended Thinking(blog.google ↗)
    264comments
  8. Negativland, Culture Jamming, and the Art of Making Something New(blog.archive.org ↗)
    19comments
  9. Building a Linux GPU Driver for the M4 Mac Mini in One Month(codyho.dev ↗)
    179comments
  10. Datamimic – don't let your coding agent invent its own test world(github.com/rapiddweller ↗)
    5comments
  11. Recreating Voodoo Graphics and a Late-1990s Gaming PC on an FPGA(nand2mario.github.io ↗)
    30comments
  12. German Rheinmetall open-sources its Battlesuite connected weapon system protcol(rheinmetall.github.io ↗)
    73comments
  13. Why I'm still bearish on LLMs after Navier-Stokes(dank.systems ↗)
    232comments
  14. A software thing I built: GPS on a 25MHz 486-SX(vcfed.org ↗)
    6comments
  15. Intelligence per Watt: Measuring Intelligence Efficiency of Local AI(arxiv.org ↗)
    2comments
  16. We got admin access to Baseten's production GitHub(strix.ai ↗)
    153comments
  17. Saving Jet Fuel(marksblogg.com ↗)
    42comments
  18. MartyPC – A Cycle-Accurate IBM PC/XT Emulator(github.com/dbalsom ↗)
    1comments
  19. Show HN: Capsule – Single-file web apps that save their data into SQLite(withcapsule.app ↗)
    135comments
  20. Learning to solve hard problems in RL for LLMs by never giving up(mnoukhov.github.io ↗)
    4comments
  21. An interactive world map of the stories cultures have told(sunnyguha.com ↗)
    5comments
  22. Let's make quality the norm again(forbrukerradet.no ↗)
    390comments
  23. Chopping up books when they're physically too big(mattkirkland.com ↗)
    156comments
  24. Jean-Pierre Serre turns 100(st-andrews.ac.uk ↗)
    24comments
  25. Suspected sabotage causes major Netherlands rail disruption(bbc.com ↗)
    426comments
  26. The Beauty of Roundabouts(gruhn.me ↗)
    75comments
  27. WangNet – 1.8 MB, zero-dependency Numberwang adjudication in 11 languages(github.com/graafhenk ↗)
    56comments
  28. Stay discoverable in search while disallowing AI training(cloudflare.com ↗)
    40comments
  29. Sierra digital cameras on the Apple II(colino.net ↗)
    5comments
  30. The Inference Hardware Revolution of 2026(ieee.org ↗)
    16comments

Intelligence per Watt: Measuring Intelligence Efficiency of Local AI

32 pointsby 1d agoarxiv.org
2 comments
13m 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.

5m 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.