there does not have to be, it is just that introducing vulnerabilities is sort of what you should expect from an llm and that they do it less often on topics with more training data is quite astounding already. unsure how to word this properly but it just seems like a technical thing from a glance

idk this case appears to be exactly the same. makes sense that it would be trained on primarily chinese data which has these things appear more rarely

saw this while listening to a pluggnb album with jrpg "aesthetics" what a coincidence

i mean yeah but thats not really what im talking about, just saying that llms can be very useful for researching stuff if you treat them as the tool they are and not a facts machine i guess

hell yeah

HI BUNNY

awww you can do it!!!

space heatress x time heatress. their relationship? the riemann metric

i agree though the one unlikely thing to consider is that theres a small chance theyre explicitly threatening him behind closed doors. dont think it Is the case but doesnt seem implausible either anymore

ah. ne wer anders

to be fair this is not entirely a new guy. usually they just like classical music (not just classical music. music from the classical period instead of trap

who!!

magical girl transformation #maybe?

the difference is the same between being outside or inside of a lambda really! you cannot really avoid these rewrites though, you just use the other tactics i suppose

theyre quite great for very specialized searches where the hallucination is not more than a mild issue since youre going to read the literature its linking too anyways

you just cant rewrite within binders sadly, unless you use conv or sometimes simp_rw