Live data from Hacker News

Viewing profile — mhjkl

mhjkl

HN member
Joined
Fri, Aug 01, 2025, 2:37 PM UTC
HN karma
31
Public activity
29 items

About mhjkl

No profile information was provided.

Recent public activity

  1. comment
    Comment #49257809

    If you adjust for writing speed, this model is already more efficient than humans. It can run at 233 tps on a 5090 which is roughly 10k words per minute compared to human typing sp…

  2. comment
    Comment #49257501

    This will make more money for Claude from individuals unofficially acting as meat puppets in more inefficient workflows

  3. comment
    Comment #49257141

    All big LLMs already visibly watermark all their text with easy to detect annoying phrases and turns of speech that everyone is already sick of hearing. Why do AI companies keep ma…

  4. comment
    Comment #49257046

    I’ve seen Chinese open weights models say “can’t use this if you’re in Europe” in their licenses, so I doubt they would invest too much into complying

  5. comment
    Comment #49198575

    You must not use any open source software, because much of it comes with a whole paragraph of ”IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES O…

  6. comment
    Comment #49077067

    If you’re on AMD, use llamacpp with the vulkan backend. For the 35B I get around 115 tps without mtp and 150-170tps with it. Also the pcie slot you use affects it

  7. comment
    Comment #48867129

    I'd say Anki has more friction that writing physical cards though

  8. comment
    Comment #48852937

    How do you make these compressed prompts like that

  9. comment
    Comment #48736359

    This comment would go hard in a sci-fi novel 5 years ago

  10. comment
    Comment #48256153

    During the dot com boom they said data centers consumed 8% of US electricity. It was later proven to be a lie, created with fake studies by coal companies to trick the US into buil…

  11. comment
    Comment #48244343

    Reminds me of the “Has The Turing Test Been Passed” website. It says no, but if you read on they cite “The relatively minimal funding allocated to AI research” as one of the reason…

  12. comment
  13. comment
    Comment #47891480

    Antis: "If AI is so useful, where are the AI shovelwares? Where are the AI open source contributions? It's all hype" 6 months later: Matz used Claude and now Ruby runs 86 times fas…

  14. comment
    Comment #47817472

    Reminds me of the Enron dataset

  15. comment
    Comment #47787859

    They didn’t need to wrap it because it’s modular arithmetic so the result after casting to int is the same regardless of wrapping behavior. 4294990000 after wrapping is 22704 and 4…

  16. comment
    Comment #47787758

    But the TSTMP_GEQ macro casts the difference to int, so any number above the signed integer limit (about 2 billion) becomes negative and the comparison returns false as they said

  17. comment
    Comment #47663419

    OP had no problem with pointers prior to trying C++. I think there is a case to be made that C(++) makes pointers unnecessarily confusing and there is no real disconnect between un…

  18. comment
    Comment #47577683

    You can remove and replace the switches too

  19. comment
    Comment #47577646

    What exactly broke? I had this problem recently and thought the keyboard was done for, but turns out you can also replace the switches (not keycaps) and that solved the issue in my…

  20. comment
    Comment #47558768

    Couldn't he just subcontract it out to whoever he needs to make his chips? Doesn't need to do anything himself

  21. comment
    Comment #47558751

    Afaik sodium batteries are much safer than li-ion and already in mass production. Unless Donut scales up really quick I think it will be more viable to just use sodium batteries fo…

  22. comment
    Comment #47554756

    There's a big difference between things like Metaverse, social media and AI. The former are mainly entertainment/communication products that rely on network effects, so if people d…

  23. comment
    Comment #47507237

    What RAM? OpenAI booked the silicon wafers, they can print anything they want on them. I wouldn't call them "far behind" on hardware when OpenAI are actively buying Cerebras chips.…

  24. comment
    Comment #47175644

    Because as we all know the EU would never try using AI for mass surveillance /s

  25. comment
    Comment #47119947

    Might as well just encode the Unix timestamp at creation into proquints.