Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Unlocking the Brain's Potential, with Gül Dölen [video](youtube.com ↗)
    discuss
  2. Where Yesterday's Decisions Run Today(medium.com/gurvinder372 ↗)
    discuss
  3. AgentLane: Git-native coordination for coding agents(github.com/thetrueshags ↗)
    discuss
  4. Ask HN: How long before we will have a major AI driven cyberincident?
    discuss
  5. Bringing Correctly Rounded Math to Production with LLVM-Libc(devblogs.microsoft.com/cppblog ↗)
    discuss
  6. When the fractional part of a float fixes your shader(crocidb.com ↗)
    discuss
  7. Otis – a minimal AI agent that runs local models out of the box(triangllabs.ai ↗)
    1comments
  8. Researchers have created a mouse with a partly human brain(economist.com ↗)
    discuss
  9. Duotone icon library with 1k icons and 250 flags for React with RTL support(github.com/andrsrxn ↗)
    discuss
  10. Show HN: Nullsec – server-blind encrypted paste(pntr.dev ↗)
    discuss
  11. Future of computing is here – OmaVision on Omarchy [video](youtube.com ↗)
    1comments
  12. Paramount 'Leaks' Study Saying California Will Suffer If Its Merger Is Blocked(techdirt.com ↗)
    1comments
  13. Mockfreeli – paste a YouTube link, get 5 license-free songs that sound like it(mockfreeli.org ↗)
    discuss
  14. The US Gov Graph – a complete map of the federal government(civlab.org ↗)
    1comments
  15. What If You Could Do It All Over?(newyorker.com ↗)
    1comments
  16. Permission to Live (In Japan)(buttondown.com/whatever_jamie ↗)
    1comments
  17. SpaceX's next Starship launch is Sept. 22, aims to reach orbit for first time(space.com ↗)
    discuss
  18. The End of Verygoodsoftwarenotvirus.ru(verygoodsoftwarenotvirus.dev ↗)
    discuss
  19. Asking Authors About Their Own Papers(medium.com/tmlrorg ↗)
    discuss
  20. Next-gen gravitational wave detectors could spot the first black holes(phys.org ↗)
    discuss
  21. US interest rates raised for first time in three years(bbc.com ↗)
    discuss
  22. My website charged AI agents a penny per page. I watched Claude pay it(suganthan.com ↗)
    1comments
  23. Show HN: Helping agents break out of their sandboxes in a monitorable way(sparrowsystems.co ↗)
    discuss
  24. Encouraging Deception in Compaction Summaries(alignment.openai.com ↗)
    discuss
  25. vLLM: Jev-like mode for the DiffusionGemma model(github.com/vllm-project ↗)
    discuss
  26. DeepSeek's Parent Company(high-flyer.cn ↗)
    discuss
  27. The Flask Mega-Tutorial, Part I: Hello, World(miguelgrinberg.com ↗)
    discuss
  28. Typesafe.ai Jev Open Source Alternative Qwen-2.5-1B-RLCD(huggingface.co ↗)
    discuss
  29. AI Debt vs. Treasuries(ft.com ↗)
    1comments
  30. "WeWorm": Zero-click worm could have taken over all WeChat accounts(heise.de ↗)
    discuss

Reverse-engineered Jev-like model

56 pointsby 5h agogithub.com
8 comments
1h agoHN ↗

No question OSS is amazing, but this video is a satire at best. It doesn't take much attention to see the results on right vs. left side are significantly different.

Jev is not interesting if it's not "smart", a 1B param model is most definitely not smart.

1h agoHN ↗

I like it! I suspect Jev may have more going on under the hood, but I like the idea of efficient universal transformers

32m agoHN ↗

Any diffusion model is potentially a Jev in disguise: https://github.com/vllm-project/vllm/pull/57250

Runs ~0.2s per decision on my DGX Spark.

  10/10 programming language detection
  9/10 human language detection
  10/12 unit magnitude comparison

All incorrect answers are marked with low-P.

It (DiffusionGemma with the Jev mode) can also solve an ASCII maze.