Reply
  • ·
    1 reply

    finally a good thread title

  • ·
    1 reply

    @Hndrxx_Free sab is jealous

  • Orangutan
    https://twitter.com/nocontextscats/status/2089353553991373239

    this video is an icon

  • Orangutan
    https://twitter.com/cuteeanimals0/status/2089608237779697931

  • CutiePieHole

    Thanks but its not true

    Not really rude, at least

  • ·
    1 reply
    frenchpress

    Yep just asked Claude the same question without memory. Same bs. It just does extra thinking to spend tokens. I think it’ll take over the bs jobs just fine but there is no AI overlord coming

    Whatre u trying to get it to do and which model

    Opus 5 can be a little oblivious and way too verbose, but Opus 4.8 is pretty rock solid to me as far as interpreting instructions and executing effectively. If you know how to prompt well then Opus 5 is better tho in my experience.

    Also assuming you're working locally via Claude Code or something like that, theres lots of little tricks and strategies to minimize token spend while getting better results. Mostly has to do with setting up skills and sorting your memory / context.

    Also if ur using an api key, anthropic can be really finnicky about token caching, so u gotta make sure what ever medium you're using Claude through supports caching for anthropic models. Sometimes you can wire up a proxy to catch your requests to and from anthropic and apply the proper headers to enable caching. Caching saves you, no exaggeration, like 90% on token costs.

  • ·
    1 reply
    Orangutan

    Whatre u trying to get it to do and which model

    Opus 5 can be a little oblivious and way too verbose, but Opus 4.8 is pretty rock solid to me as far as interpreting instructions and executing effectively. If you know how to prompt well then Opus 5 is better tho in my experience.

    Also assuming you're working locally via Claude Code or something like that, theres lots of little tricks and strategies to minimize token spend while getting better results. Mostly has to do with setting up skills and sorting your memory / context.

    Also if ur using an api key, anthropic can be really finnicky about token caching, so u gotta make sure what ever medium you're using Claude through supports caching for anthropic models. Sometimes you can wire up a proxy to catch your requests to and from anthropic and apply the proper headers to enable caching. Caching saves you, no exaggeration, like 90% on token costs.

    give it up claude. you will never rule over us

  • Awful new hot sauce i got tainted my whole salmon. Pissed . Beyond belief

  • kanye getting really rich again from shows allegedly...might be a cultural force again in 2028

  • ·
    1 reply
    frenchpress

    give it up claude. you will never rule over us

    its actually pretty funny because most of these AI models have hit a major plateau, since they're all built off the exact same model outlined in a paper from a few years back. Without getting too technical, they've reached the point where linear improvements require exponential increases in parameter count, data size and computing power.

    So while things may get a little better in the near future as they find workaround with hybrid setups and whatnot, nothing is gonna get jaw-droppingly better until someone finds a totally new way to do AI lol

  • LLMs are basically just roided-out text prediction machines that they gave access to the breadth of human knowledge, then rolled the dice on different tuning combos millions and millions of times until they found one that works better than the last

  • ·
    1 reply
    Orangutan

    its actually pretty funny because most of these AI models have hit a major plateau, since they're all built off the exact same model outlined in a paper from a few years back. Without getting too technical, they've reached the point where linear improvements require exponential increases in parameter count, data size and computing power.

    So while things may get a little better in the near future as they find workaround with hybrid setups and whatnot, nothing is gonna get jaw-droppingly better until someone finds a totally new way to do AI lol

    thank you claude, finally you agree with me. Shall I subscribe to the pro max plan now?

  • ·
    1 reply
    Like a Camillimane

    1 million per video but u have to watch and can't mute

    how many vids are u getting through

    @Prince @KodakSpice

    https://www.youtube.com/@​theswiftologist/videos

    ive seen 1

  • ·
    1 reply
    CutiePieHole

    When gta 6 online drops can we get married in game?

    depends

  • frenchpress

    thank you claude, finally you agree with me. Shall I subscribe to the pro max plan now?

    The Orangutan account actually utilizes the Qwen3.6Plus model from AlibabaCloud to generate its posts

  • my arm hurts

  • Spice Bones

    depends

    Ok

  • Would hndrx pop my balloon

  • ·
    1 reply

    stack bread

  • ·
    2 replies

    Please let me get pregnant in gta online !

  • Like a Camillimane

    finally a good thread title

  • CutiePieHole

    Please let me get pregnant in gta online !

    knuts busted

  • Spice Bones

    ive seen 1

    Damn.

  • ·
    1 reply
    CutiePieHole

    Please let me get pregnant in gta online !

    just watched the leaked vids...yeah kh4 wiping this s*** down revelation wiping this s*** down...

  • I need to drink some more.