@goblinodds In my experience despite the context limit bump LLM output degrades pretty rapidly beyond the ~200k mark (depending on a lot of factors). Fable has disappointed me sometimes on non-tech-related stuff though
@Foiblaet As I've gotten older I've realized many or perhaps even most people's world models ossify fairly early in life and rarely if ever update.
I wonder if this has always been the case or if it's a byproduct of wealth/tech leaving us increasingly insulated from reality
@wylfcen We're decolonizing English!
I wonder, how hard would it be to reconstruct a neo-old English with more modern words borrowed or inferred from still-extant north germanic languages?
@nosilverv I've been thinking this often with respect to code generation, with people acting like human programmers reliably generated good code pre-AI. The 30 year old pile of spaghetti I maintain at my job is testament to the wrongness of this notion
@marshmarigold I walked right by them to take the elevator to the parking garage and they didn't ask any questions; guess they were slacking last night
@earnestpost Insane to me that you even have to make such considerations when they've been letting anyone in for years. You should be able to just take an IQ test