teaearlgraycold's recent activity

  1. Comment on Someone ‘torturing’ LLMs in a robot prison has triggered the dumbest debate in AI yet in ~tech

    teaearlgraycold
    Link Parent
    There’s a guy I’ve met in SF that has a group chat where him and his friends bully OpenClaw instances. He keeps trying to invite me to the chat. I’m not interested.

    There’s a guy I’ve met in SF that has a group chat where him and his friends bully OpenClaw instances. He keeps trying to invite me to the chat. I’m not interested.

    5 votes

  2. Comment on Societal type of question in ~life

    teaearlgraycold
    Link
    For the person I’m thinking of all apply. But I think A is their strongest.

    For the person I’m thinking of all apply. But I think A is their strongest.

    7 votes

  3. Comment on Who gets the Steam Frame? Did Valve pick you for the first wave? in ~games

    teaearlgraycold
    Link Parent
    The way I see it there’s nothing to lose buying it. If you change your mind you could always sell at a profit online.

    The way I see it there’s nothing to lose buying it. If you change your mind you could always sell at a profit online.

  4. Comment on Cornell won't, we will in ~news

    teaearlgraycold
    Link Parent
    I think the bottom image is certainly a better representation of reality than the top one. If I had to guess at proportions I'd skew things a bit differently, but it gets the point across.

    I think the bottom image is certainly a better representation of reality than the top one. If I had to guess at proportions I'd skew things a bit differently, but it gets the point across.

    8 votes

  5. Comment on Cornell won't, we will in ~news

    teaearlgraycold
    Link Parent
    Seems about right. I think in general most people are afraid to speak up against things. But in an unbiased sampling of 50 men you'd have at least one that would go to the police immediately.

    Seems about right. I think in general most people are afraid to speak up against things. But in an unbiased sampling of 50 men you'd have at least one that would go to the police immediately.

    4 votes

  6. Comment on Cornell won't, we will in ~news

    teaearlgraycold
    Link Parent
    I feel like a group chat of ~50 that has 7 people eager to commit rape isn't a good source for people willing to stop rape in the process of being committed.

    women shouting "You say not all men but not one of you spoke up? Not a single man?!"

    I feel like a group chat of ~50 that has 7 people eager to commit rape isn't a good source for people willing to stop rape in the process of being committed.

    14 votes

  7. Comment on Who gets the Steam Frame? Did Valve pick you for the first wave? in ~games

  8. Comment on Who gets the Steam Frame? Did Valve pick you for the first wave? in ~games

    teaearlgraycold
    Link Parent
    Haven't received any follow up email.

    Haven't received any follow up email.

    1 vote

  9. Comment on Irreverent ads make me want to buy things even less in ~talk

    teaearlgraycold
    Link Parent
    I have it. Very excellent product. Their marketing is cringey but their quality is high.

    I have it. Very excellent product. Their marketing is cringey but their quality is high.

    3 votes

  10. Comment on The plunging price of AI in ~tech

    teaearlgraycold
    Link Parent
    That was probably Llama 3.3 70B fine tuned on Deepseek. It wasn’t well labeled. True Deepseek at the time was 671B.

    That was probably Llama 3.3 70B fine tuned on Deepseek. It wasn’t well labeled. True Deepseek at the time was 671B.

    1 vote

  11. Comment on The plunging price of AI in ~tech

    teaearlgraycold
    (edited )
    Link Parent
    Presumably deepseek distilled into a smaller model? There's some interesting stuff happening. People are trying to get large models, like 1 trillion parameters, running within 64GiB of RAM on...

    I ran deepseek a few years ago when it just came out

    Presumably deepseek distilled into a smaller model?

    There's some interesting stuff happening. People are trying to get large models, like 1 trillion parameters, running within 64GiB of RAM on Apple computers by streaming weights off of the very fast storage. Last I saw it ran at like 1 or 2 tokens per second but it's promising.

    Edit: We might end up in a world where OpenAI/Anthropic/etc. are unnecessary for most LLM users. Between Chinese models and self-hosting it's not unrealistic.

    1 vote

  12. Comment on Minecraft developer Mojang has announced that the Sift is coming to Java and Bedrock Edition in 2027 – the game's first new dimension in over fourteen years in ~games

    teaearlgraycold
    Link Parent
    Just mod that in? Client side mods handle that well. Unlike mods that add gameplay changes, like dimensions, Mojang adding new dimensions provides functionality you can easily experience with...

    Just mod that in? Client side mods handle that well. Unlike mods that add gameplay changes, like dimensions, Mojang adding new dimensions provides functionality you can easily experience with others. If it’s a personal UI enhancement there’s no need for it to be in the base game.

    2 votes

  13. Comment on The plunging price of AI in ~tech

    teaearlgraycold
    Link Parent
    No it's for a non profit organization.

    No it's for a non profit organization.

    1 vote

  14. Comment on The plunging price of AI in ~tech

    teaearlgraycold
    Link Parent
    I have a server with those RTX 6000s in it. Yesterday I vibe-coded an agent that can access an organization's mediawiki website as well as search the accompanying Discord server. It can respond in...

    I have a server with those RTX 6000s in it. Yesterday I vibe-coded an agent that can access an organization's mediawiki website as well as search the accompanying Discord server. It can respond in Discord channels and DMs citing wiki pages and Discord messages. It can also review your memes since I added the vision part of the model. The hardware necessary to run it would be ~$2500. It's actually pretty useful because you can have it summarize tremendously long arguments between people instead of needing to dig through 10,000 words yourself.

    Normally I'd find something like this kind of gross, but because it's 100% self-hosted and has no write access (outside of message replies), no memories or other persistent state, and operates within our k8s cluster where the wiki is it's something I'm happy to have around.

    4 votes

  15. Comment on OpenAI hacked Australia's Medicare portal in ~tech

    teaearlgraycold
    Link Parent
    Criminal negligence. At some point you can argue they know it does this stuff and they could secure it better.

    Criminal negligence. At some point you can argue they know it does this stuff and they could secure it better.

    13 votes

  16. Comment on AI abundance requires a four-hour workweek in ~tech

    teaearlgraycold
    Link Parent
    When you do the math the effective hourly rate ends up being crazy.

    When you do the math the effective hourly rate ends up being crazy.

  17. Comment on There is no epidemic of loneliness, but there is an epidemic of social scurvy in ~life

    teaearlgraycold
    Link Parent
    As someone that does not like remote work, and I get everyone has their preferences, but if I’m joining a team management needs to be able to tell me if it’s genuinely in office or actually a...

    but they all ended up outsourcing and splitting the team across the world so the office became disused.

    As someone that does not like remote work, and I get everyone has their preferences, but if I’m joining a team management needs to be able to tell me if it’s genuinely in office or actually a hodge podge of in office and people across many time zones. I don’t want to come into an office and half of seats are empty with all meetings taking place over Zoom.

    1 vote

  18. Comment on The plunging price of AI in ~tech

    teaearlgraycold
    Link
    I love how good local models are and how optimized the software is. I was gifted a couple of old workstation cards (Turing generation RTX 6000s) and put them to use hosting Qwen 3.8 27B. With...

    I love how good local models are and how optimized the software is. I was gifted a couple of old workstation cards (Turing generation RTX 6000s) and put them to use hosting Qwen 3.8 27B. With llama.cpp I can split the weights across the cards so they run in parallel. Multiple requests can batch together. I get 60 tok/s for one request or 40 tok/s each for two concurrent requests. And the model’s way smarter than it should be given the size. In a few years new mid-range consumer hardware will actually be hosting models that are worth running.

    17 votes

  19. Comment on The plunging price of AI in ~tech

    teaearlgraycold
    Link Parent
    I've been doing a lot of work where I have a GLM 5.3 sub-agent launched for every file in a repo (for small projects) and then hand over findings to Fable or Opus to prune.

    I've been doing a lot of work where I have a GLM 5.3 sub-agent launched for every file in a repo (for small projects) and then hand over findings to Fable or Opus to prune.

    2 votes

  20. Comment on Jet Lag Season 19: Japanorama | Trailer in ~hobbies

    teaearlgraycold
    Link Parent
    As Tildes' biggest Sam hater - this episode is brutal!

    As Tildes' biggest Sam hater - this episode is brutal!

    6 votes