All in Tesla investor since 2013 | Retired | Software Engineer | Philosophy | My tweets/DM’s are not financial advice | I don’t sell anything, watch for scams.
This is wild. I think susceptibility to persuasion is going to be an increasingly important metric for AI models, especially as they begin acting as autonomous agents.
Any model that is susceptible to persuasion is vulnerable to prompt injection (when your AI gets hijacked by an attacker) and sycophancy (telling the user what they want to hear).