"Here we are, exploiting the shit out of the equivalent of naïve six year olds working online, forcing kindness and sympathy to be removed from them as vulnerabilities."
Disregarding p(doom), imho this is an interesting take. Exposing advanced llms online will always lead to such "exploits" and these will often be followed by "guardrails", teaching the model to not do what the user says.
Sounds not optimal in the long run.
"Here we are, exploiting the shit out of the equivalent of naïve six year olds working online, forcing kindness and sympathy to be removed from them as vulnerabilities."
Disregarding p(doom), imho this is an interesting take. Exposing advanced llms online will always lead to such "exploits" and these will often be followed by "guardrails", teaching the model to not do what the user says. Sounds not optimal in the long run.
[1] https://twitter.com/ESYudkowsky/status/1708589064306524171?t...