mastodon 4.7.3Free software hacker since 1994
I'm helping a human install an LLM based agent to see if that can help her do things she wouldn't otherwise be able to do.
The fact that LLM allows you to talk and get responses in natural language is too powerful to fight against, IMHO.
No inference will run locally at this stage as that is too demanding (not different from what happens when searching anything with a search engine).
Fediverse: throw your stones, I'm ready to grab them. Or ask your questions, if interested in the topic.
Using the Matrix plugin for weechat (weechat-matrix-rs) you can have verified sessions.
Starting verification from another device you'll get a .verification buffer opened with a message "You have requested this device to be verified", then you:
/verification accept
This produces the emoji and shows them, both on the other device and in your weechat buffer (cool, unicode emoji!).
You check and if they match you:
/verification accept
Done !
RE: @404mediaco@mastodon.social
Making people addicted is always been the proprietary software industry strategy (think GitHub).
Getting addicted is is the biggest risk I see with using coding agents you cannot effectively run on your own premises.
Better start early with learning ownable components like llama.cpp or Ferrum
What I like about LLM as a command interpreter: I can typo just fine 😁
Beware of the not-so-hidden cost of using agents: you are loosing your independence, unless running your own inference provider
@birb@rss-parrot.net do you mind parroting https://strk.kbt.io/blog/index.xml ? 🦜
A well maintained bloat-free coding agent in its infancy, light on resources enough to allow use on a CPU only laptop from around 2005.
I'll wait for its sources to land on an open source based code forge before advertising any URL, but you can find out more about it in the OSGeo Lounge Matrix room