pull down to refresh
33 sats \ 0 replies \ @UCantDoThatDotRugged 22 Feb \ on: Google Gemini leaked its woke prompt injection process. news
'Someone' and no source. And no context. This could be the continuation of a conversation that makes interpretation of this final response much different. I believe that Gemini and others do this, and even as described is possible, but this is no technical explanation of how it works, just an out of context portion of a GPT generated response, which we need to learn to be skeptical of.
https://m.stacker.news/17103
=============
https://m.stacker.news/17104