2 ms·
I was just playing a board game with a friend, and we spent a while looking through the (fairly complicated) rules. At some point, I had the bright idea to feed
by stavros 3y ago
I was just playing a board game with a friend, and we spent a while looking through the (fairly complicated) rules. At some point, I had the bright idea to feed the manual to Claude, so we could ask questions about it.
It all went great, Claude even gave references, until it hallucinated an innocuous but significantly game-breaking rule ("rest lets you heal one health for two stamina").
This just reinforced my lack of trust in LLMs, and I would sure as hell not ask an LLM what drugs were safe or not.
- NobleLie 3y agoI'd think the application would be used more for understanding the psychedelic experience. Not about safety. There is such a wide tome of experiences - there are going to emerge similarities to a knowledgeable AI that can load it all up at once. I'm very curious what patterns can emerge with up to date language models.
- _a9 3y agoI dont think erowid content would be good to tell safety of a drug, probably more what type of experience you'd get using a drug at certain doses. What would be interesting would be training using the data from PsychonautWiki[0] which has all sorts of good data about drug information & safety (and experiences) [0] https://psychonautwiki.org/wiki/Main_Page https://psychonautwiki.org/wiki/Main_Page