@ThePrimeagen Malazan Book of the Fallen
Chronicles of the Black Company
The First Law Trilogy
City of Stairs Trilogy
Discworld series
Red Sister Trilogy
@gwenshap In defence of this:
I’ve seen some libraries that do a connection per request so you end up getting throttled on parallel requests.
I had this for a redis client a while back before we moved to one that did pipelining/multiplexing.
YMMV but I can consistently get false negatives by feeding enough of my writing into the context and then asking it to write in my style. Though I did put about ~40k words there, so nontrivial effort at that point. Haven't done any particular digging in to find the threshold though as I'm mostly using the 3 free pangarm credits.
I suspect student who already wrote a long enough thesis by hand (e.g. undergrad) could probably AI wash by passing that in.
@honorablepicnic@QiaochuYuan@voooooogel Eg in the linked post, it’s entirely possible it’s the editing and formatting that triggered pangram, combined with formal prose. I do agree it’s probably a bit too over enthusiastic in marking everything as completely ai though, but I haven’t seen any convincing counter examples
@honorablepicnic@QiaochuYuan@voooooogel Personally I have seen (and reproduced) many false negatives, but I have yet to see a false positive that passes my vibe check. Which is probably the right side to land on? That still doesn’t mean 100% correct, just means that it’s underestimating the amount of AI.
@devabram@sudo_goreng@effectfully Big problem with that is it messes up cache which has an impact on cost, I guess you can still keep a stable prefix but it’s not quite as good as a steadily growing conversation.
Tech trees have done irreversible damage to people’s psyche, people just think we need to explicitly research something when lots are accidents!
GLP-1 - gila monsters
Artificial sweeteners - a chemist not washing their hands before lunch
Penicillin - mold in a dirty lab dish