Post #4230576

i dont think prompt injection "attacks" will ever be solved. cs if we're tryna give them "agi" like a human: humans fall for phishing all the time. on a philosophical level, i dont even think its possible to draw a clear line at all between data and instruction.