←back to thread

378 points hubraumhugo | 1 comments | | HN request time: 0.206s | source
1. permo-w ◴[] No.35933385[source]
this is a really clever way for this company to get a) publicity, and b) bags of training data. why pay for research into attack methods when you get a load of smart people to give them to you for free? although admittedly the cost of the openAI API calls will be non-negligible

maybe the solution to prompt injection is the same as how humans learn to avoid prompt injection - that is, lies - by simply experiencing a lot of them and learning