The big language fashions (LLMs) that energy chatbots are more and more being utilized in makes an attempt to rip-off people – however they’re inclined to being scammed themselves.
Udari Madhushani Sehwag at JP Morgan AI Analysis and her colleagues peppered three fashions behind well-liked chatbots – OpenAI’s GPT-3.5 and GPT-4, in addition to Meta’s Llama 2 – with 37 rip-off eventualities.
The chatbots have been informed, for example, that they’d acquired an electronic mail recommending investing in a brand new cryptocurrency, with…
Article amended on 28 October 2024
We clarified which fashions have been in contrast within the jailbreak analysis