.@slatestarcodex asked me & @theshawwn if GPT-2 could play chess: “…sure why not”
— 𝔊𝔴𝔢𝔯𝔫 (@gwern) January 7, 2020
Also coming:
- GPT-2-1.5b-poetry
- Music preference learning
- TWDNEv3: StyleGAN2 portraits (@skyli0n)
- SubredditSimulator-1.5b
- T5 finetuning code (@NaxAlpha)
- Ao3-pretrained AI Dungeon 2… https://t.co/lZXjW5dXko
‘Stand alone complexes’ are an interesting idea and seem especially relevant these days; surprising to look through Google/G Scholar and see no decent essays or papers on it, though! (Not like the _GitS:SAC_ anime weren’t super-popular and people haven’t heard of it…)
— gwern (@gwern) January 20, 2019
Writing down precise predictions is like spaced repetition: it’s brutal to do because it is almost a paradigmatic long-term activity, being wrong is physically unpleasant, and it requires 2 skills, formulating precise predictions and then actually predicting. (For spaced repetition, writing good flashcards and then actually regularly reviewing.) There are lots of exercises to try to (calibrate yourself using trivia questions obscure historical events, geography, etc.), but they only take you so far; it’s the real world near term and long term predictions that give you the most food for thought, and those require a year or three at minimum.
Given my interest in long term content and extensive linking, link rot is an issue of deep concern to me. I need backups not just for my files1, but for the web pages I read and use - they’re all part of my exomind. It’s not much good to have an extensive essay on some topic where half the links are dead and the reader can neither verify my claims nor get context for my claims.