2 ms·
Exactly. Recently, this has been reminding me of all the hand wringing about "prompt injection vulnerabilities" - like e.g. when you have an LLM summarizing ar
by TeMPOraL 3y ago
Exactly.
Recently, this has been reminding me of all the hand wringing about "prompt injection vulnerabilities" - like e.g. when you have an LLM summarizing articles for you, or going over your inbox, and a malicious actor sneaks in a well-crafted bit of text that makes that LLM return you false information, or forward all your e-mails to attacker's address, etc.
Well guess what, ads are exactly such prompt injection attacks, but on humans.