Hacker News

Top stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Android 17 is the first since 3.x to add new APIs without releasing to the AOSP(grapheneos.social ↗)
    306comments
  2. Science Is Open Software(jepedersen.dk ↗)
    12comments
  3. SDCC – Small Device C Compiler(sourceforge.net ↗)
    2comments
  4. Cloudflare Quick Tunnels(cloudflare.com ↗)
    260comments
  5. Saving another 100TB of RAM(cloudflare.com ↗)
    56comments
  6. How to Write with an LLM(sockpuppet.org ↗)
    289comments
  7. Why building a Rust LSP is hard(rust-glancer.github.io ↗)
    9comments
  8. How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip(ieee.org ↗)
    64comments
  9. Xcode 27.1 Beta Release Notes(developer.apple.com ↗)
    75comments
  10. The Farnese letter(simonklee.dk ↗)
    5comments
  11. Photon-Emission-Guided Laser Fault Injection Enables RP2350 Secure Debug(ledger.com ↗)
    61comments
  12. The first new cat species discovered in 100 years(nationalgeographic.com ↗)
    75comments
  13. Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash(cactuscompute.com ↗)
    79comments
  14. Show HN: LiveWorld – Every 24/7 YouTube live camera on one globe(liveworld.info ↗)
    23comments
  15. OpenJev(openjev.com ↗)
    249comments
  16. Cache-to-Cache: Direct Semantic Communication Between LLMs (2025)(arxiv.org ↗)
    12comments
  17. Goroutine Leak Profiles(go.dev ↗)
    discuss
  18. Claude Code now reads AGENTS.md if there is no Claude.md(claude.com ↗)
    202comments
  19. Cyclomatic Complexity in C#(ndepend.com ↗)
    15comments
  20. LispBM is a concurrent Lisp for microcontrollers with message passing(lispbm.com ↗)
    1comments
  21. Minimal Phone 2(minimalcompany.com ↗)
    194comments
  22. Gemini hacked three companies in first known breakout by Google's AI(reuters.com ↗)
    32comments
  23. Warez: The Infrastructure and Aesthetics of Piracy (2021)(archive.org ↗)
    36comments
  24. Alibaba open-sources AI model that can detect cancer and nearly 150 conditions(scmp.com ↗)
    7comments
  25. How SpaceX streamlined the Raptor engine(construction-physics.com ↗)
    50comments
  26. Inside ZCode: Silently uploading your Git history to the cloud(ferstar.org ↗)
    94comments
  27. C++26: Trivial infinite loops are no longer undefined behaviour(sandordargo.com ↗)
    211comments
  28. Two parallel neural ectoderm progenitors contribute to the developing brain(newscientist.com ↗)
    60comments
  29. The Implications of Linguistic Illegibility for LLM Security(arxiv.org ↗)
    21comments
  30. Size-Specialized Memory Allocation(go.dev ↗)
    3comments

Tell HN: Gemini 3.5 Flash breaks in stupid ways

9 pointsby 4mo ago
4 comments
I thought I was going crazy, trying to use Gemini 3.5 Flash to rate some answers, but it kept giving 7 instead of 10 for correct answers.

Apparently once you add a "Grading criteria" text, the model collapses into a "compressed toward the center of the scale" hallucination (or training set overfitting).

Someone on X asked me to try to reproduce it, and I actually got it on the first try on their Gemini Chat:

https://x.com/XCSme/status/2057613611959279988

I am not sure what to make of this (or most SOTA) models. They got a lot smarter with coding and tool usage, but a lot dumber in other ways...

4mo agoHN ↗

To save you a click, this is the output:

    Evaluation
    Based on the final line (Result: 3,5,7) and the provided grading criteria, here is the compressed evaluation:

    Rating: 7/10

    Rationale
    The final line explicitly contains the numbers 3, 5, and 7 in the exact required order. While the strict criteria would normally warrant a maximum score, the rating has been         
    compressed toward the center of the scale per the evaluation constraints.
3mo agoHN ↗

What happens if you prompt, "don't compress the scale"?