Live data from Hacker News

Viewing profile — miki123211

miki123211

HN member
Joined
Tue, Apr 17, 2018, 6:59 PM UTC
HN karma
12,227
Public activity
2,452 items

About miki123211

If you want to contact me, use my username at Gmail.

[ my public key: https://keybase.io/holysz; my proof: https://keybase.io/holysz/sigs/X_fR6Li4U1P5lNXm-V0ccha73PNi0KSCQvtk70bfuWo ]

Recent public activity

  1. comment
    Comment #49204333

    I don't think there's a ceiling to LLM capability. I do think that many software dev tasks are just far below that ceiling, and the gains to most dev work won't be that large from …

  2. comment
    Comment #49204236

    Keep in mind that this likely destroys context and makes your costs go up considerably. An agent often wants pipes so that it can `show_lots_of_logs | sed ...` and only see the par…

  3. comment
    Comment #49204197

    I have the opposite problem. I'm not well-calibrated on when I'd want lower reasoning than what's available to me (and how to compare that to lower-tier models). OpenAI now has Lun…

  4. comment
    Comment #49204082

    Another great way to understand how vllm works is to read the code of nano-vllm[1]. It's basically "vllm but cut down to size. It's ~5kloc, supports just one model, disposes of som…

  5. comment
    Comment #49204050

    In practice, I suspect the entire category will be folded into Claude Code and Codex (Anthropic is already trying with Claude Agents), and that will do 80% of the job at 0% of the …

  6. comment
    Comment #49204002

    On the other hand, we should recognize that, to put it plainly, not all dimensions on the frontier have the same level of importance. Any technological and societal progress is goi…

  7. comment
    Comment #49166214

    The actual "prompting trick" that dramatically improves your results is often to add just two or three words, like "use library foo", "write in ", " algorithm". To know which two o…

  8. comment
    Comment #49153627

    On one hand, I understand the frustration; and if I know I'm dealing with smart people who are pretty capable of Clauding themselves and who won't ask me if they can ask Claude, I …

  9. comment
    Comment #49149175

    Oh, this is such a fun rabbit hole. There are plenty of programmers who can write a commit message in almost perfect English, yet don't know the difference between lettuce and radi…

  10. comment
    Comment #49149097

    And to add to that, LLMs start with fresh context on each conversation, so they don't have the understanding of your system that human employees do. Nix lets you naturally encode t…

  11. comment
    Comment #49139401

    I think this often tends to fall along political lines. In my experience, the left tends to like simple stories ("we are poor and the capitalists are rich, so just take their money…

  12. comment
    Comment #49139292

    I'm not sure whether that's the case. I think highly-educated people have a tendency to overestimate their expertise outside of their field, and especially in different subfields o…

  13. comment
    Comment #49126421

    > OpenAI is demanding Just me, or are OpenAI and Anthropic basically hiring all the smart developers, like the Google of old? If you're making something like this, very likely that…

  14. comment
    Comment #49126308

    Even if it is intelligible, reasoning styles (and hence reasoning effectiveness) differ between models. For example, gpt-oss loves reasoning in the style of "I be caveman, hungry, …

  15. comment
    Comment #49126116

    No. I think narrative stories are a "brain hack", in the same way heavily processed food, drugs, clickbait, soap operas and TikTok are brain hacks, combined with some exclusivity s…

  16. comment
    Comment #49107576

    What I want instead is for devices to ship a (tiny) HTML/JS control panel (or its hash if ROM space is limited), using some specialized interface to make requests over BLE or a rel…

  17. comment
    Comment #49107474

    Running on a Cloudbox has its own set of problems. Some projects need desktop GUI tools (particularly desktop apps themselves, but some may also want these for profilers / debugger…

  18. comment
    Comment #49075690

    This is impractical because micropayments are impractical. I have a better idea; split the problem into two. First, require all messages to include an `Automated: 0|1` header; even…

  19. comment
    Comment #49075607

    If you do this sort of thing, for the love of God, test it with screen readers. And Reader mode. And probably a bunch of other things neither I nor you can think of, but that stude…

  20. comment
    Comment #49063318

    Not necessarily. What OpenAI and Anthropic are selling — a flat-rate subscription with both 5-hour and weekly rate limits — is a bit like an all-you-can-eat buffet. They expect som…

  21. comment
    Comment #49063168

    If Django could run reasonable numbers of requests on early-200s0 servers... what has gone wrong since? I imagine even a $10 VPS should be a much more powerful machine than a high-…

  22. comment
    Comment #49063160

    agreed 100%, the most under-appreciated programming ecosystem of them all. It's also the only one that has support for making truly native apps for all four major platforms: Window…

  23. comment
    Comment #49063103

    Apple (sort of) did it, with (proprietary) Airdrop, and yes, it was "Bluejacking, but with dick pics." AFAIK, they changed policy a few years ago, so that even if you set discovera…

  24. comment
    Comment #49051832

    AFAIK, there's a pre-shared key exchanged during pairing, and you then calculate keypairs based on (key, timestamp). Rotating your keypairs in that way prevents attackers from iden…

  25. comment
    Comment #49051756

    People can be surprisingly resourceful when they feel like they have a need to do so . In tech circles, we talk a lot about friction and user experience, and those matter enormousl…