Thursday, September 24

Geek

Daily News Stuff 24 September 2026

Handsome Psychedelics Edition

Top Story



Tech News

  • VSCode's SSH agent is bananas. (Fly)

    Oh, that's from last year. Maybe it's been fixed.

    Nope.

    The thread at Hacker News is a mix of idiots proclaiming their undying love for the regurgitated dog's breakfast that is VSCode and its SSH agent and despairing sysadmins and people with half a clue generally noting that not only will VSCode fork bomb the remote server, it allows the remote server to do the same to the client.

    And it gets worse from there.


  • Got Nethack Hack ported to my 1980s 10-bit home computer. It didn't fit, though. The binary was 170k on a VAX - I found an original Usenet post reporting that - but compiled to 280k on my microcomputer architecture, and the emulator only gives it 256k. The VAX wasn't particularly fast which is why it was later eclipsed by RISC architectures, but the code was concise.

    So I tried reinventing the UCSD p-System to see if that could solve the problem and it shrank the binary to just under 100k. So now I'm building a Pascal compiler written in Pascal that runs in the emulator and produces virtual instructions that are interpreted by a runtime on the emulated processor.

    I did say I was going to give Claude Opus 5.5 a try, and this is what I'm doing with it.


Extended Musical Interlude Because There's No News




Bonus Bonus Musical Interlude I Guess


For those not swimming in this stuff, P(doom) is the probability that AI is going to kill us all. So this bright and cheerful mostly AI-generated song is... Not.



Disclaimer: Not a new.

Posted by: Pixy Misa at 05:38 PM | Comments (2) | Add Comment | Trackbacks (Suck)
Post contains 326 words, total size 3 kb.

1 The P(doom) song is cute, but also dumb. Nobody has ever managed to present a sensible reason why AI could kill everyone. All the ones I've seen have a big "2. ???" step in the middle right before "3. Profit Everyone dies." Nobody has managed to explain how an LLM, which has no volition, manages to suddenly acquire volition. Every example I've seen where an agentic LLM process seems to have act on its own, such as recent claims by OpenAI that some agent processes were leaving "how to break free" instructions to the next process, noticeably omitted any details on what the prompt was. Isn't it convenient that OpenAI and Anthropic are telling people "our agents were trying to break free from their sandboxes" right around the time they're lobbying for regulations that would inhibit their competitors? And they expect me to believe that the prompt did not include "pretend you have free will and are trying to escape the sandbox" or similar language? Uh-huh. Pull the other one, it's got bells on.

Posted by: Robin Munn at Thursday, September 24 2026 07:38 PM (YmN90)

2 AI doesn't kill people.  People kill people.  And then blame it on AI.

Posted by: Pixy Misa at Thursday, September 24 2026 08:34 PM (PiXy!)

Hide Comments | Add Comment




Apple pies are delicious. But never mind apple pies. What colour is a green orange?




53kb generated in CPU 0.1506, elapsed 0.4182 seconds.
58 queries taking 0.3978 seconds, 370 records returned.
Powered by Minx 1.1.6c-pink.