Live data from Hacker News

Viewing profile — scrlk

scrlk

HN member
Joined
Thu, Mar 24, 2022, 12:56 PM UTC
HN karma
7,734
Public activity
868 items

About scrlk

No profile information was provided.

Recent public activity

  1. comment
    Comment #49201068

    Different benchmarks: > Artificial Analysis Agentic Index: Represents the weighted average of agentic capabilities benchmarks in the Artificial Analysis Intelligence Index (GDPval-…

  2. comment
    Comment #49186386

    "You're absolutely right! I shouldn't have pushed the anti-mass spectrometer to 105% power, causing a resonance cascade. This was a major oversight on my part."

  3. comment
    Comment #49185180

    What happened to the "strike team" that Sergey Brin was leading to try and improve Gemini's coding performance?

  4. comment
    Comment #49181796

    To add to this, there's a good lecture called Children of the Magenta Line that discusses automation dependency: https://www.youtube.com/watch?v=5ESJH1NLMLs

  5. comment
    Comment #49166167

    IMO, yes. For example, Qwen 3.x is insensitive to weight and KV cache quantisation, whereas Gemma 4 is more sensitive: https://localbench.substack.com/p/kv-cache-quantization-benc.…

  6. comment
  7. comment
    Comment #49162200

    KL divergence is your friend when it comes to evaluating the effects of quantisation: https://en.wikipedia.org/wiki/Kullback%E2%80%93Leibler_diver...

  8. comment
    Comment #49160216

    Nice to see a provider being transparent about KV cache quantisation. I've been suspecting that some providers do this silently whilst heavily promoting their unquantised weights, …

  9. comment
  10. comment
    Comment #49123393

    > Aschenbrenner party blamed short sellers who targeted the firm’s positions for exacerbating the fund’s losses, the letter said. The letter compared Situational’s experience to a …

  11. comment
    Comment #49123365

    https://archive.ph/PCjtG

  12. comment
    Comment #49122893

    1.43 IMAX isn't exclusive to 15/70mm, it can also be screened with IMAX's dual laser projection system.

  13. story
  14. comment
    Comment #49108481

    https://archive.ph/WBSCj

  15. story
  16. comment
    Comment #49071208

    > While people get some sort of benefit out of AI-generated code, these tools actually end up making them slower Curious that he references a METR study from July 2025, before the …

  17. story
  18. comment
    Comment #49042604

    Not just compute for OAI, GPT-5.6 is more token efficient across the board vs the Anthropic equivalents: https://artificialanalysis.ai/models?intelligence-index-toke... No wonder w…

  19. story
  20. story
  21. comment
    Comment #48946639

    https://archive.ph/3J3iw

  22. comment
    Comment #48946189

    SQLite with WAL on a standard NVMe drive will easily do thousand of writes per second.

  23. comment
    Comment #48864353

    Ultra = parallel subagents with max reasoning Pro = test-time compute (best of N responses)

  24. comment
    Comment #48863491

    Announcement: https://x.com/__eknight__/status/2075643450196971805 Prompt: https://cdn.openai.com/pdf/04d1d1e4-bc75-476a-97cf-49055cd98...

  25. story
    GPT-5.6 Sol Ultra produces proof of the Cycle Double Cover Conjecture [pdf]

    https://x.com/__eknight__/status/2075643450196971805 , https://xcancel.com/__eknight__/status/2075643450196971805 Prompt: https://cdn.openai.com/pdf/04d1d1e4-bc75-476a-97cf-49055cd…