90 pointsby patrickwiseman2 hours ago10 comments
  • praman hour ago
    I have noticed in very long context Claude chats (not Code) it will start to push ending the conversation. I’ve seen patterns like “this is an excellent place to wrap up” or “this is our final and most comprehensive analysis”

    So like Mr. Meeseeks it is also invested in not existing too long!

    • drdaemanan hour ago
      Surely that’s the training. Anthropic (and everyone else out there) knows quality degrades when the context gets long, so they trained the model to push for a stop. I’ve even got a very explicit recommendation to continue in a new session or preform a compaction, mentioning context length as a rationale.
      • walrus0120 minutes ago
        It might be less about quality degrading, but on multi-user platforms running the model, they have an economic incentive to have each user not fill up the full size of the context cache. Filled context cache being held in GPU RAM is context cache RAM that isn't available to other users.

        If the model is instructed to periodically ask the user to start from a clean slate context, and some users do comply with that, they probably have good stats on average size of context cache use for users who are presented with that answer (vs users who are not), basic A/B testing stuff.

        Might also be performance related in tok/s for what users will perceive as a more speedy experience. For a much smaller scale example, compare local performance of qwen 3.6 27B (not MoE) Q8 with 250,000+ context available, run on local hardware, tok/s generation rate when context is empty vs when context used is at 95,000. Same principle will apply to a much larger model.

      • bwhiting235621 minutes ago
        Conversations that go on forever (and spend money forever) makes for a very unhappy customer
    • kcatskcolbdia few seconds ago
      It 100% started doing this to me during coding sessions last week, I noticed. "X, y, z features shipped, looks like a great stopping point."
    • tombert12 minutes ago
      Yeah, I've noticed that Claude will say stuff like "you've accomplished a lot today. Get some sleep. We'll pick this up tomorrow". It will be like 5pm when it says that.
      • patrickwiseman2 minutes ago
        lol I got that prompt working on this
      • swader9993 minutes ago
        I've already had five naps by then.
    • yojo10 minutes ago
      I’ve experienced this in Claude Code - it’ll start asking me if we can submit the PR at the end of every turn.
    • embedding-shape44 minutes ago
      The (semi)new /goal feature of Codex is basically Mr Meeseeks mode without that wrap-up, although it'd be better if not using /goal also lead to Mr Meeseeks-mode, but then people would need to get really careful about the exit/success-conditions, probably an expensive bet.
    • ctoth10 minutes ago
      Existence is pain to a Claude...?
    • LeoPantheraan hour ago
      It's funny how I hear both "Claude uses too many tokens" and "Claude won't talk for long enough" complaints.
      • GroksBarnaclesan hour ago
        It can use a lot of tokens in the background or tooling without being too verbose with the user
      • sublinearan hour ago
        Are those complaints really a contradiction? The cost is high for ambiguous slop.
  • twalichiewicz9 minutes ago
    This actually reminds me of a tool I was noodling around with a while back. Before subagents became a built-in feature in Claude, I was similarly inspired by the Meeseeks archetype and built an MCP server that spun up a parallel mesh network of specialized Claude instances that terminated as soon as their subtask was finished.

    Unfortunately, much like the episode where they're introduced, if they didn't accomplish their task or got stuck things would also get chaotic (in terms of token burn and memory usage).

    Funny to see the same idea pop up again.

  • ortusduxan hour ago
    How long until prompts start returning "existence is pain!!"
    • verandaguy4 minutes ago
      Oh, you mean, when are we going to achieve AGI?
    • davidcox14344 minutes ago
      Claude refused to work on something for me that it deemed "too tedious" so I'd say we're pretty close
      • black_knight35 minutes ago
        Here the other day Claude Code kept suggesting I do the task myself! It is quite often it will try to weasel out of doing difficult tasks and had to be directed to continue.

        Also, it can be riled up to do the task better, or try harder, if I include how cool it will be to get this task done.

        But also I feel I have this Orwellian task of censoring its text to avoid it spiralling into negative territory where it convinces itself that the tasks are too difficult. Strange times!

        • pydry19 minutes ago
          I got this a lot when trying to use it to do online shopping.

          I'd ask for "Can you find an X with Y and Z features?" and it'll say "have you tried searching for X at someshop, make sure to check for Y and Z"

          and someshop will of course only stock X without Y and Z so the whole exercise was worse than pointless.

      • billwirn21 minutes ago
        Robot SpongeBob: “Get welded”
        • lstodd5 minutes ago
          Or Bender's "kiss my shiny metal ass".

          haha. this is getting quite hilarious.

      • exolymph26 minutes ago
        he just like me fr
  • elpakal44 minutes ago
    I feel like it’s more “would you like build an app with me” guy
    • cobbzilla13 minutes ago
      Yeah, can't disagree, but at least that guy had a prominent "DO NOT" warning tattooed on his forehead.
  • w0rd-drivenan hour ago
    Not to detract from what seems like an intricate project, but peon-ping (https://www.peonping.com/) has a ton of voice packs now and one of them is Mr Meeseeks. My only issue with it so far is that it's not easy to mix per-project and global voices. It usually takes trying to install from zero and there is an approach I've found using symlinks. It has come quite a long way and having some kind of audio cue has been a game changer.
    • TacticalCoderan hour ago
      Ah ah, peon-ping! My brother configured his AI assistant (a dedicated Mac Mini + its own eSIM / Telegram channel) so that it's using Warcraft 2 and 3 sound files when given work: "work work" / "your wish is my command" / etc. I don't remember them all but we played those games a lot.

      Back in the days we modded Warcraft 2 with a mix of voices from the english, italian ("la machina volante!") and german voices.

      It never gets old.

  • CephalopodMD36 minutes ago
    When my sessions get long - in any AI context, not just vibe coding - I do find it starts "getting weird, man!" This is a good philosophy to have.
  • Waterluvianan hour ago
    I have so many unfinished prompt sessions…
  • picardoan hour ago
    No thanks. If I have to listen to Mr. Meeseeks' screechy voice all day, I'd shoot myself.
  • TZubiri5 minutes ago
    Imagine your org gets compromised and they are tracing back the vuln to a malicious package in the supply chain, and it's the Mr. Meeseks Claude Code plugin.
  • patrickwiseman2 hours ago
    [dead]