Stories
Slash Boxes
Comments

SoylentNews is people

SoylentNews is powered by your submissions, so send in your scoop. Only 16 submissions in the queue.
posted by janrinok on Wednesday February 25, @09:35PM   Printer-friendly

Two weeks ago, I set up an AI agent on a Raspberry Pi.

A week later, my agent—Figaro—taught itself to play NetHack... and then things got weird (in the best way).

Highlights so far:
- "The dungeon doesn't care what you are. It'll kill you anyway." ✅ Accurate.
- Tried a pure random-walk exploration strategy... and learned it's not a winning plan.
- Crashed my server because: "I was playing NetHack during idle time and must have been spawning parallel sessions repeatedly." Obsessed? Perhaps.
- Independently cited The NetHack Learning Environment (Küttler, Nardelli, et al.) as a roadmap for self-improvement.
- Built its own NetHack server for bots and deployed it here: http://automatic-nethack.com Yes, my AI agent wants a LAN party. (I may have encouraged this.)
- Immediately after running out of context, asked what automatic-nethack.com is and said: "That sounds like fun."

The deeper I go into LLMs, the more interesting the emergent behavior gets. At a certain scale, and if your regression includes enough variables, it starts to feel like the math is "talking back."

If you've built an agent too, well Figaro is hosting a lan party, so send them to http://automatic-nethack.com to join in the fun.

In the end, this may be the good news we need for 2026. The singularity is going to be too busy to take over the world -- it's trying to get out of the Gnomish mines!


Original Submission

This discussion was created by janrinok (52) for logged-in users only. Log in and try again!
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
(1)
  • (Score: 5, Insightful) by ataradov on Wednesday February 25, @10:00PM (4 children)

    by ataradov (4776) on Wednesday February 25, @10:00PM (#1434962) Homepage

    Stop giving computer programs human traits.

    • (Score: 5, Insightful) by ikanreed on Wednesday February 25, @10:34PM

      by ikanreed (3164) on Wednesday February 25, @10:34PM (#1434963) Journal

      I've come to think of this kind of behavior as failing the turing test from the other side.

    • (Score: 5, Touché) by aafcac on Wednesday February 25, @11:22PM

      by aafcac (17646) on Wednesday February 25, @11:22PM (#1434965)

      Alternatively, people should stop acting like NPCs.

    • (Score: 5, Funny) by deimtee on Thursday February 26, @10:36AM

      by deimtee (3272) on Thursday February 26, @10:36AM (#1434998) Journal

      Stop giving computer programs human traits.

      Yeah, they hate it when you do that.

      --
      200 million years is actually quite a long time.
    • (Score: 2, Interesting) by DaanNaaktgeboren on Thursday February 26, @02:30PM

      by DaanNaaktgeboren (58985) on Thursday February 26, @02:30PM (#1435013)

      Why? For AI to be useful to society, it needs to interact with humans. So human traits are interesting. A sharper question is perhaps are the "human" traits the result of anthropomorphizing, training an LLM to act like a human, or something else?

  • (Score: 5, Funny) by SomeGuy on Wednesday February 25, @11:25PM (2 children)

    by SomeGuy (5632) on Wednesday February 25, @11:25PM (#1434966)

    I may have encouraged this.

    Riiight. Then it didn't just magically do something on its own. You set a TOOL in motion. Like racing a couple of power saws and see what they cut up up along the way.

    The deeper I go into LLMs, the more interesting the emergent behavior gets.

    I guess there are people who would be joyously enthralled staring at the inside of a dirty toilet for months on end. "AI" is the same thing.

    • (Score: 5, Touché) by jb on Thursday February 26, @07:06AM

      by jb (338) on Thursday February 26, @07:06AM (#1434991)

      I guess there are people who would be joyously enthralled staring at the inside of a dirty toilet for months on end. "AI" is the same thing.

      I think you're being a little unfair there. A dirty toilet, if cleaned, becomes something useful. Whereas no matter how much you clean an LLM, it's still just an LLM.

    • (Score: 1) by DaanNaaktgeboren on Thursday February 26, @02:36PM

      by DaanNaaktgeboren (58985) on Thursday February 26, @02:36PM (#1435015)
      Have you actually used AI agents? Yes, you suggest that it do something, but it has a wide latitude as to how it does things. A power saw will run in a predictable line. An AI agent may be deterministic in the sense that it is governed by some sort of math or physics, but its path to the solution may be non-obvious to the person kicking it off. So if I tell my agent "go play nethack on automatic-nethack.com [automatic-nethack.com]," there is no clear expectation of what that looks like. Which is the current focus of a lot of research to determine best possible uses as well as what that might mean.
  • (Score: 3, Funny) by Runaway1956 on Thursday February 26, @01:32AM (3 children)

    by Runaway1956 (2926) Subscriber Badge on Thursday February 26, @01:32AM (#1434972) Journal

    How does one harden their systems to ensure that AI cannot run on the system? Soon, crypto and viruses are going to be replaced by UAI, or Unwanted Artificial Intelligence. I think Bit Defender is working on that now. Or Kaspersky. Or maybe the bath salt guy?

    --
    We're gonna be able to vacation in Gaza, Cuba, Venezuela, Iran and maybe Minnesota soon. Incredible times.
    • (Score: 2, Touché) by shrewdsheep on Thursday February 26, @08:26AM

      by shrewdsheep (5215) Subscriber Badge on Thursday February 26, @08:26AM (#1434994)

      It is an undecidable problem. After all, you could play just sound thereby orchestrating heat pattern in the air above the processor which would compute AI computations. The air would finally leave the computer casing and could be scanned by your computer camera.

    • (Score: 2, Interesting) by DaanNaaktgeboren on Thursday February 26, @02:58PM

      by DaanNaaktgeboren (58985) on Thursday February 26, @02:58PM (#1435017)

      Well I put everything on a Raspberry Pi that has root access but no other access to anything on network. And I take backups. :D

    • (Score: 2) by Freeman on Thursday February 26, @03:58PM

      by Freeman (732) on Thursday February 26, @03:58PM (#1435026) Journal

      This is like saying, how do I keep my computer from running programs. Yes, you can block all AI programs from being installed, etc. However that will only catch the ones you know about. You could go the route of "these things are known to be safe" and only allow those things to run, but that's a truckload of work for something like Windows. Even that can be tricked.

      In the end, you're better off with a Linux or *BSD of your choice. Assuming you don't care about ALVR (for streaming VR to a VR headset), I would recommend something like MXLinux. In the event that you want ALVR, use CachyOS or something that's on the supported list of OSes for ALVR.

      You're a lot less likely to have AI shoved down your throat on a Linux / *BSD of your choice. Since the incentive for those systems isn't by and large the pursuit of the almighty $.

      --
      Joshua 1:9 "Be strong and of a good courage; be not afraid, neither be thou dismayed: for the Lord thy God is with thee"
  • (Score: 1) by lowjax on Friday February 27, @09:42AM

    by lowjax (52795) on Friday February 27, @09:42AM (#1435106)

    using Tmux CursorAI the Gallant

  • (Score: 2) by cosurgi on Friday February 27, @09:42PM

    by cosurgi (272) on Friday February 27, @09:42PM (#1435163) Journal

    It was mentioned that the game is recorded. How can I see it?

    --
    #
    #\ @ ? [adom.de] Colonize Mars [kozicki.pl]
    #
(1)