AI After An Apocalypse

Hello! I hope you enjoyed the talk.

If you have feedback, questions, suggestions, or if you would like to talk about the talk, you can email aaaa@psimyn.com.

TLDW

1. Lots can go wrong with LLMs, or cloud stuff more generally
2. You can run good LLMs on consumer hardware, and you probably already have some
3. There are fun and interesting things to explore if/when internet is not available.

Links

Slides are here

Things mentioned:

• Linux Cowboy book

• LM Studio

• Unsloth Studio

• Odysseus

• llama.cpp now at llama.app

• llm

• llama-swap

• pi

• opencode

• llmfit

• walkie talkie webapp

• WIA Foundation Radio License practice test

Post-talk meta stuff

Overall it went well. People seemed to enjoy it, maybe even learned something. I got about 70% of what I wanted to cover, but it could have used more practice or polish. Forgot to mention some things, forgot some jokes. Overall very nervous and rusty.

I had written some speaker notes. Ended up not being near my laptop, decided to stand out near the mirror display away from the lectern. That and the countdown clock both trapped me a little, I found myself looking a bit much at the mirrored screen rather than audience. Also just feeling frozen in one spot, had anticipated pacing or something.

Under-prepared timing. I had tried talking for 18 minutes and it went mostly ok. The plan was mostly 4 minutes each for intro/why/Local LLMs/network. Then the clock was counting down instead of up so I completely lost track of what I should be up to when.

Main screw up was last section, both the remote dev envs and then home cloud stuff was very muddled.