Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Evaluating S2S Model Quality with Human Preference Data(withdavid.ai ↗)
    discuss
  2. "Start with a Monolith" Was Good Advice. AI Is Changing That(medium.com/scalar-engineering ↗)
    discuss
  3. Snap launches $2,195 Specs smart glasses(yahoo.com ↗)
    discuss
  4. DeepSeek-v4.1 Flash: Pushing the Limits of KV Cache Compression(zartbot.github.io ↗)
    discuss
  5. What regulatory capture actually looks like(marginalrevolution.com ↗)
    discuss
  6. Tin: full-text search for Postgres(planetscale.com ↗)
    discuss
  7. Part-human part-mouse brain developed in science breakthrough(bbc.com ↗)
    discuss
  8. Show HN: FastRecall, ultra-cheap memory across AI models(fastrecall.ai ↗)
    discuss
  9. Michael Burry slams OpenAI, Anthropic for 'self-serving' calls to slow AI(nypost.com ↗)
    discuss
  10. Spanish data watchdog publicises first AI agent-linked data breach report(reuters.com ↗)
    discuss
  11. Xcode-Project-Format(github.com/apple ↗)
    discuss
  12. Flowchart: How pixels become an Apple Reference Image(claude.ai ↗)
    discuss
  13. Supply Chain Compromise of Korean-Language Windows 11 Installation Media(logpresso.com ↗)
    discuss
  14. Pangram – AI detector for text and images(pangram.com ↗)
    discuss
  15. Migrating the GitHub Copilot Runtime to Rust, Using Copilot(github.blog ↗)
    1comments
  16. Game UI Database(gameuidatabase.com ↗)
    discuss
  17. Found a B2B billing stack for my agency that doesn't feel like 2010(cordhq.app ↗)
    discuss
  18. Pro UI: native grade components for pro software(pro-ui.dev ↗)
    discuss
  19. Page Shield ML caught 4 storefront malware campaigns scanners missed(cloudflare.com ↗)
    discuss
  20. Why the Postpandemic Tech Bust Sent Billionaires to Trump(wired.com ↗)
    discuss
  21. Hacker puts 'full redundancy' code-hosting firm out of business (2014)(pcworld.com ↗)
    discuss
  22. OpenAI Discloses Six New Incidents of ‘Concerning’ A.I. Behavior(nytimes.com ↗)
    8comments
  23. Math challenges a 2k-year-old story about Parthenon's optical illusions(phys.org ↗)
    discuss
  24. Could Your Chatbot Be Stealing Ideas from You?(nytimes.com ↗)
    1comments
  25. Weeping whales: Stillborn humpback whale grieving documented(phys.org ↗)
    discuss
  26. I gave my agents a heartbeat(mlsystemsri.com ↗)
    discuss
  27. Will AI Replace Your Doctor? Dr. Zeke Emanuel and AMA President Dr. John Whyte(youtube.com ↗)
    1comments
  28. A Large Database Does Not Mean Large Shared_buffers(keithf4.com ↗)
    discuss
  29. Size-Specialized Memory Allocation(go.dev ↗)
    discuss
  30. What's Scarier Than Agents Taking over Internet? CEO Cartel Trying Take over AI(fractalsofchange.substack.com ↗)
    1comments

Reverse-engineered Jev-like model

76 pointsby 6h agogithub.com
10 comments
3h 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.

2h agoHN ↗

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

1h 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.

39m agoHN ↗

Out of curiosity and semi unrelated — why do so many of these projects with customized encoder-decoder setups use earlier Qwen versions like 2.5 and 3 and not the smallest 3.5? Purely the few 100m params, or something else in the latter’s arch or pretraining?

9m agoHN ↗

In my experience if you tell Claude to port LLM-like stuff without explicit steering for versioning, it will default to the most popular thing for this in its training window to reduce errors. 3.5 is outside its training data.