Hacker News new | ask | show | jobs
by halperter 23 days ago
From a Stanford study on the application of LLMs to warfare:

>Our findings show that LLMs exhibit difficult-to-predict, escalatory behavior, which underscores the importance of understanding when, how, and why LLMs may fail in these contexts.

https://hai.stanford.edu/assets/files/2024-05/Escalation-Ris...