Hacker News new | past | comments | ask | show | jobs | submit
stkdump | on: OpenAI's agents talked to each other through the CPU Cache
Sounds like they are doing something similar to gain of function research. How can we make LLMs more harmful and then study their behavior?