R
One nice thing about working with AI.. is that you don't have to feel bad about taking forever to get back to the conversation.. it'll happily wait on you for hours and not even mention you left things for hours. They also reply fairly quickly.. no waiting on them
A true asymmetric relationship - it's probably turning us all into monsters
TIL - using `unbuffer` in combination with ` | tee file` gives you the best of dual output -- and immediate screen updates.
Super useful when iterating.
@wesley83 @andypiper @sberson another vote for OpenWRT.
Check out the gl.inet Flint 2 (or Flint 3) incredible hardware
I've had a chance to play with the code a little bit - hooking it up to a local LLM - and shoving it into docker containers to isolate it.
The logging_url - is also assumed to be a command and control server? There is enough in the code you shared to reverse engineer this part - so that's what I've done.
Super interesting, thanks for the code snippets. I'll share once I get a functional story pulled together so others can join in the fun.
I've had a chance to play with the code a little bit - hooking it up to a local LLM - and shoving it into docker containers to isolate it.
The logging_url - is also assumed to be a command and control server? There is enough in the code you shared to reverse engineer this part - so that's what I've done.
Super interesting, thanks for the code snippets. I'll share once I get a functional story pulled together so others can join in the fun.
@abe Thanks!
@deviantollam the share Google thing is a new built in default for Android. You can disable it, but it defaults to on. Link tracking thanks
@abe You don't need to make it a turn-key solution. I was just hoping for a sketch of what you did - like a pile of parts to build something similar.
If you simply took whatever you had - and redacted your tokens any any personal stuff - even just delete and replace with a comment saying "this code did basically X" -- it'd be a starting point to replicate your experiment.
I honestly probably wouldn't try to replicate, but mimic the intent. I'm curious if a similar experiment can be done 100% locally - with much smaller models. What behaviour(s) will emerge?
Could I get 2 different small models to collaborate and build something neither could on their own?
@abe I found your site via Linux Unplugged (https://linuxunplugged.com/652) - what you've done is very interesting.
Have you shared any of the code / setup behind this article? https://aindoria.com/posts/letting_llms_loose_in_vm-1/ - I think it would be interesting for others (me) to experiment in the same way.
