I have found my most major frustration with LLMs: The lack of context. Small context windows and no memory beyond a few thousand tokens is utterly useless for anything I would want to use an LLM to do. Are there any models that actually have useful memory? If so, is is possible to self-host them using hardware that doesn't cost thousands of dollars?

Language aside, this is the proper attitude. View quoted note →

I think my biggest issue with open source AI that you can run on hardware that doesn't cost more than most make in a month is the small context windows. This is stupid. So, so stupid. Everything needs MORE context, not less. This is super frustrating and I really don't understand why LLMs are built this way. At all. I am, however, starting to learn more about how LLMs are run, and what it takes to make them performant, but I'm still a decade behind and I'm not sure I want to catch up.

I am *very* glad I insisted that my dad and I stack wood last year. The prediction of a long, cold winter has born out to be totally accurate and it's still holding on, but won't outlast the wood pile. Stacking sats is all fine and dandy, but, it isn't enough. Humans cannot live by sats alone. It would be wise to not forget that. #randomthoughts #morningwood #stackwood

Good morning! This is the only way:

Good morning. This is me every day:

I support this man's bid for king! View quoted note →

A tale of two cows:

Sour cream and salsa is one of the best combos ever. #brunch #sourcream #salsa