suprjami

11 ms·

1. 181.▲
   by suprjami 11mo ago

   First time reading The Register huh?
2. 182.▲
   by suprjami 1y ago

   You're right about "reasoning". It's just trying to steer the conversation in a more relevant direction in vector space, hopefully to generate more relevant output tokens. I find it easier to conceptualize this in three
3. 183.▲
   by suprjami 1y ago

   God forbid having to learn how to use something useful. Don't ever take up programming.
4. 184.▲
   by suprjami 1y ago

   Thanks for the summary. American dishwashers don't have their own heater? All dishwashers I've seen in Australia only have cold water supply.
5. 185.▲
   by suprjami 1y ago

   In that case I suggest you buy cheaper desktop cards instead of a H200. Two or three 5090s will let you run decent models at very good speed.
6. 186.▲
   by suprjami 1y ago

   I somehow have the memory that there was an extremely narrow time window where the speedup was tangible and quantifiable for Gentoo, as they were the first distro to ship some very early gcc optimisation. However it's open source softw
7. 187.▲
   by suprjami 1y ago

   Fun tinkering but I don't find it very compelling to get root access by becoming the root user. A more exciting usage is pairing this with qemu userspace emulation, so you can uncompress a filesystem from another architecture. You can
8. 188.▲
   by suprjami 1y ago

   If you have ~$25k to buy a H200 then don't buy one. Rent them out much cheaper and keep renting newer models when your H200 becomes an outdated paperweight. Assuming you ran inference for the full working day, you'd need to run yo
9. 189.▲
   by suprjami 1y ago

   Correction, if you're going to get a Mac then get a Max or Ultra, with as much memory as possible, the increase in RAM bandwidth will make running larger models viable.
10. 190.▲
    by suprjami 1y ago

    Qwen 3 Coder is a 30B model but an MoE with 3B active parameters, so it should be much faster for you. Give it a try. It's not as smart as dense 32B for general tasks, but theoretically should be better for the sort of coding tasks fro
11. 191.▲
    by suprjami 1y ago

    Copy of https://github.com/kelseyhightower/nocode which has been around a lot longer with additional humour in the Issues section.
12. 192.▲
    by suprjami 1y ago

    Now you've duped everyone with the wrong headline, let's discuss what we thought it was about. I bet ancient people saw themselves as the pinnacle of civilisation, much like we do now. I'm sure Romans were sitting there with
13. 193.▲
    by suprjami 1y ago

    But they do. They're not talking about AGI, they're talking about venture capital funding.
14. 194.▲
    by suprjami 1y ago

    And as this post shows, it doesn't.
15. 195.▲
    by suprjami 1y ago

    Agree, any Max or Ultra should walk all over this thing, and has the advantage of many years of already-working software. Apple benchmarks: https://github.com/ggml-org/llama.cpp/discussions/4167
16. 196.▲
    by suprjami 1y ago

    And a 14B model running at 22tg/s means you won't be using that 128G RAM for inference either.
17. 197.▲
    by suprjami 1y ago

    So I can spend thousands of dollars to have an unstable training environment and inference performance worse than a US$200 3060. Wow. Where do I sign up?
18. 198.▲
    by suprjami 1y ago

    > LINAGORA is a French open-source software publisher, a pioneer in digital sovereignty, and a leader of ethical, user-centered solutions
19. 199.▲
    by suprjami 1y ago

    One of my favourite webpages. Have used it countless times over the years.
20. 200.▲
    by suprjami 1y ago

    iiuc the high price is mostly from the high bandwidth memory. (which isn't actually that high bandwidth compared to actual GPUs)
21. 201.▲
    by suprjami 1y ago

    Hasn't this existed for many years? I used to use Firefox profiles to separate work and personal over 10 years ago, nowadays I just use a different browser.
22. 202.▲
    by suprjami 1y ago

    > home-baked container implementation > everything runs as root EasyOS - An experiment in speedrunning a CVE
23. 203.▲
    by suprjami 1y ago

    I ran TTRSS many years ago after Google Reader went away. There was one particular feature I really liked that people kept making forum threads asking about, and the maintainer would post angry rants in reply. Eventually he said if one more
24. 204.▲
    by suprjami 1y ago

    For some reason I read this as "Tritanic" and "Bitannic" and thought there was a cool series of ships with a mathematical naming convention. Oh well.
25. 205.▲
    by suprjami 1y ago

    Depending on what you want to do, you already can. llama.cpp and other inference servers work fine on the kernel driver.
26. 206.▲
    by suprjami 1y ago

    Guess it's time to find a new launcher then. Suggestions?
27. 207.▲
    by suprjami 1y ago

    Why does it look like an image of an asteroid hitting a planet? https://stock.adobe.com/images/asteroid-hitting-the-earth-ai...
28. 208.▲
    by suprjami 1y ago

    No. We have MicroSD Express. We don't need yet another form factor.
29. 209.▲
    by suprjami 1y ago

    I have separately documented the same thing happening on our work knowledgebase. TCP is always such fun.
30. 210.▲
    by suprjami 1y ago

    I think that's unlikely to get widespread traction. Source code is not for computers, it is a way for human developers to communicate with each other. Compilers/interpreters are a consumer of that communication. Without easy commu