Old profile: luccus@feddit.de

Mastodon: luccus@chaos.social

  • 0 Posts
  • 10 Comments
Joined 2 years ago
cake
Cake day: June 28th, 2024

help-circle




  • I have mostly given up on stopping them. I my testing with local bullshit, they will eventually work out most red herings. And I imagine the bazillion parameter models will do so even faster.

    However, you can eat into the token budget, sometimes a lot, by adding sarcasm, aggressive/unusual in-lining or having some “fun” in your naming schema and descriptions (if the language allows them). Same with solutions requiring lateral thinking.

    But the best way is to not give the full story easily. Some IDEs may have toggles for stuff, the LLM would have to find in the some config file. Game engines will let you set things up in the GUI a LLM will not easily pick up while injecting the code. Anything outside the code is usually a win.



  • their software has commited multiple crimes

    Their management allowed for, and their technicians excecuted the careless set up of a Rube Goldberg crime machine. This occured after a series of other crimes and while taunting the ability of their machine to commit crimes.

    The creation of the machine also included crimes, meanwhile augmenting the ability of criminals to commit even more crimes.

    Surely they will be brought to justice! (insert Padme meme)





  • Current “AI” is actually shitty tho’.

    Any frozen weight models are architecturally unable to lean anything new, while dynamic weight models require entire datacenters to run for every instance. The scale of low-effort slop has crippled the internet as a knowledge source, and we are currently poisoning entire towns and the earth at large making up problems to solve.

    Look at Nvidia, look at Huggingface. Huggingface literally offers their AI detection right next to their AI obfuscation. Meanwhile Anthropic is constantly hyping bullshit. I just read the “J-Space” article, and holy crap, LLMs use hidden layers to encode and infer relarionships, wow. I thought this was the whole idea behind machine learning, but I guess, saying that won’t secure another billion dollars.

    The entire space is full of people without any idea what they are looking at, and with LLMs taking up all the space and money, actually good ideas, like probabalistic proteine folding, are tossed to the side.

    The only kinda good thing I see is Markdown making a comeback. And that’s just “good”, because I like Markdown.


  • This just reminded me of a classmate from my teens who once said, in all seriousness, “Hitler didn’t just do bad things”.

    And I replied, “Uh, yeah”. And in my mind, searching for something good he did, went like ‘he built the Autobahn’ immediately followed up with ‘well, that didn’t turn out to be the best idea either’, and by then the moment had kinda passed, and I couldn’t casually say “Well, actually […]”.

    Man, politics as a teenager was wild.