Hacker News

Top stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005(cryptocellar.org)
    242comments
  2. OpenAI is well positioned to fast-follow Jev(arcturus-labs.com)
    78comments
  3. Jev – a curation of Jev demos on X, tools, skills, and integrations(github.com/amal-david)
    16comments
  4. Apple has added persistent 'ads' to iOS, and it's driving users crazy(techradar.com)
    181comments
  5. 16-bit Intel 8088 chip by Charles Bukowski(allpoetry.com)
    2comments
  6. Show HN: Drop – a rootless Linux sandbox with gVisor support(droprun.sh)
    21comments
  7. Solitaire Alone Together(solitairealonetogether.com)
    3comments
  8. Writing Rust code that's fast by asking agents to make the code faster(minimaxir.com)
    discuss
  9. Can gzip be a language model?(nathan.rs)
    119comments
  10. MiMo v2.6(xiaomi.com)
    458comments
  11. I asked Meta’s Muse for its filesystem and it sent me 6.8GB(mouse.dev)
    56comments
  12. Spymarks, Not Watermarks(brand.io)
    153comments
  13. Muse, Meta's extraordinarily privileged AI assistant, has a serious 0-day(arstechnica.com)
    20comments
  14. Teleoperated Humans(jefftk.com)
    17comments
  15. Line Scan Photos from MUNI Heritage Weekend in San Francisco(lawrence.lu)
    16comments
  16. Attention is all you have(alicegg.tech)
    296comments
  17. Vacate a Drone Restriction That Criminalized Recording Immigration Agents(eff.org)
    6comments
  18. Transformers Explained Visually(poloclub.github.io)
    83comments
  19. What Sun got wrong(dtrace.org)
    378comments
  20. I said no and Apple said yes(dbushell.com)
    495comments
  21. The Economics of Open-Weight Inference(ornn.com)
    4comments
  22. Xbox continues its "reset" with dramatic restructuring(arstechnica.com)
    discuss
  23. A font that reads what you wrote(rohanadwankar.github.io)
    37comments
  24. I don't want to read what you didn't write(colinbreck.com)
    384comments
  25. AMD's random number generator can't generate a 0?(flatassembler.net)
    142comments
  26. MiMo-v2.6-Pro: Intelligence, Performance and Price Analysis(artificialanalysis.ai)
    56comments
  27. AI coding has made CI a bottleneck, so we reworked ours to keep up(linear.app)
    362comments
  28. Engineering Memory: On learning to memorize first 100 digits of pi (2024)(gregorygundersen.com)
    31comments
  29. 9 Ads per Minute: FIFA Cup 26 – "the price of the beautiful game"(bristol.ac.uk)
    196comments
  30. Divide by depth for instant 3D(gabrieloc.com)
    35comments

The Economics of Open-Weight Inference

10 pointsby 2h agodata.ornn.com
4 comments
24m agoHN ↗

The problem, IMO, with open-weight models is that you accustom to the capabilities of frontier models too quickly; and downgrading to an open-weight "frontier minus 2" or "frontier minus 3" model is often painful, since they feel way less useful than their newer closed-weights counterpart. To be honest, I don't know any companies using OW models at a large scale for their operations (agents or chat assistants).

21m agoHN ↗

I think an interesting point is that hardware as of today still has no utility value after its reported lifetime has elapsed, which prevents neolabs and smaller labs from getting older HW clusters as the banks are not willing to give out loans against them. There is no agreed upon pricing for "expired" A100 clusters or similar.

This is clearly not true, and we are starting to see compute markets, but only for rental prices/H, not for the hardware itself. I feel like there is some artificial moat being built here to stimulate sales of new hardware, because an H100 at 1/16th the price will have comparable dollar/FLOP as Vera Rubin.

10m agoHN ↗

Depends on the workload. H100 will never have the network performance of Vera Rubin. There's also token per watt, newer systems will beat the older systems.

4m agoHN ↗

Then why do people still believe that OpenAI any Anthropic have negative margins