Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Show HN: LLMs Play Chicken in Realtime (wildcardlabs.tech)
    —discuss
  2. Sonnet 5.5 scores just behind Opus 5.5 on Artificial Analysis Intelligence Index (artificialanalysis.ai)
    —discuss
  3. The Ratchet (rubick.com)
    —discuss
  4. Tech OpenAI abandons plan to release upcoming model as safety concerns escalate (cnbc.com)
    —discuss
  5. Joseph Henrich's Report on Anthropology (vanderbilt.edu)
    —discuss
  6. Mosquitoes Are a Choice (worksinprogress.co)
    —discuss
  7. Go Pace Yourself (an example of what AI films done right looks like) (reddit.com)
    1comments
  8. Electronics in U.S. homes by income [pdf] (eia.gov)
    —discuss
  9. Rogue Agent Investigation by Asymmetric Security (asymmetricsecurity.com)
    —discuss
  10. The Transparent Society (1998) (wikipedia.org)
    —discuss
  11. Duet For The End Of Math [pdf] (caltech.edu)
    —discuss
  12. OpenAI Scrapped Newest Model Release (bloomberg.com)
    —discuss
  13. Taking 'execute logging' a bit too literally CVE-2026-88771 (europa.eu)
    —discuss
  14. The Edge of California Is Falling into the Ocean (theatlantic.com)
    —discuss
  15. A diagram language where you say where things go (reladraw.github.io)
    —discuss
  16. This Battery Is Ready to Eat (acs.org)
    —discuss
  17. Show HN: Pac-Bench – How well can models one-shot a Pac-Man game? (jonclegg.github.io)
    2comments
  18. What if automating AI R&D triggers an intelligence explosion? [pdf] (casp.ac)
    —discuss
  19. I built a FIDO2 hardware key. OpenAI says it isn't hardware enough (voidnullvalue.github.io)
    —discuss
  20. Frogs Wear Bacterial Armor to Defend Against Fungus (nautil.us)
    —discuss
  21. Synthetic Aperture Radar Drone Gets Interferometric Imaging (hackaday.com)
    —discuss
  22. AI Almost Started a U.S.–China War – and No One Seems to Care (theintercept.com)
    —discuss
  23. Optimize Tetris NP-Complete Solution with AlphaEvolve (bufferof.com)
    —discuss
  24. 10 Tokens for Your Meat Brain LLM (jimmysastra.com)
    1comments
  25. Reliable AI automation done right (not with a model) (skeptical.ai)
    —discuss
  26. AI is eroding the barriers that kept biological weapons rare (statnews.com)
    3comments
  27. SBX env: consistent and shareable AI sandbox configurations (docker.com)
    —discuss
  28. Meta Poaches MongoDB CEO "Effective Immediately" (tikr.com)
    —discuss
  29. Show HN: Procedural universe generation without N-body loops (github.com/pjhkorea)
    1comments
  30. Mathathon: Old Problems, New Proofs (mathathonchallenge.com)
    —discuss

Show HN: Pac-Bench – How well can models one-shot a Pac-Man game?

3 pointsby 29m agojonclegg.github.io
2 comments
Benchmarks how well Harness+models can create a Pac-Man game from a single prompt:

“Create a Pac-Man game in a single HTML page”

Each model gets one shot — no follow-up prompts or fixes.

25m agoHN ↗

Opus 5-5 seemed like a perfect clone, with others displaying flaws in an initial look. Astra notably created a bunch of surrounding ugly crap to look at.

13m agoHN ↗

I was just coming here to say this. I looked at all of them, and Opus 5.5 at high effort seems significantly better than every other clone.

Buffered controls, different pathfinding AI for each ghost, level transitions, etc.

Gpt-5.6 sol high technically completed the assignment as well, but the gaps within the pipes used to construct the maze, the lack of pause when eating a ghost, etc made it feel far less polished.