Scott Williams 🐧 on Nostr: Truly open source models trained on public domain or fully owned data require more up ...

Truly open source models trained on public domain or fully owned data require more up front investment but also carry less foreseeable risk than other options. To that end, ARM hardware is momentarily cheap at a time where very little of anything is cheap, so buying some Ampere boxes and having it scrape your own website and data is a pretty good proposition. Whatever LLM service, don't expect it to reliably think or make decisions for you.