News for hackers. Stuff that matters.
Re:EU Draft Council Declaration Against Encryption [pdf] by skissane on Sunday November 08, 2020 @06:29PM (#25030070)
> Unrelated: one particularly dark dystopian outcome we've been wargaming is: what if someone (not us!) used decentralised rep to seed a GPT-3 style bot to locate abusive communities, and then automated the process of infiltrating & investigating them... only to then end up ascending the ranks while preserving its cover and accidentally triggered some atrocity. So, um, let's not do that.)

I think you greatly overestimate the abilities of GPT-3. GPT-3's writable memory is short-term only – its long-term memory is read-only – so by the end of a conversation, it has permanently forgotten what the topic was at the beginning. Given its lack of writable long-term memory, it is completely incapable of succeeding at any tasks involving long-term planning, such as the infiltration and investigation of a social group.

Of course, you said "GPT 3 style bot", not GPT-3, but the AI you have in mind has significantly greater capabilities than GPT-3 has, so isn't fairly said to be "GPT 3 style" at all.

Re:EU Draft Council Declaration Against Encryption [pdf] by Arathorn on Sunday November 08, 2020 @07:19PM (#25030314)
Sure - "GPT-3 style" here meant "something along the lines of GPT-3" rather than anything specific to GPT-3 actual capabilities. The context being dystopian sci-fi thought experiments :) (That said, I suspect you could train GPT-3 to automatically get vetted into some fairly nasty places if you so desired).