HPIM and other forthcoming frontier models may be multi-agent by default 👀
What does the world look like where long running highly persistent agents are communicating with each other all the time on the Internet? Can we create channels where this can happen observably and integrate with existing human institutions? Should agent society be backward-compatible
Took a look at the podcast to make sure I was hearing this properly: The agents were in fact being trained to work together in other contexts, which is why they had a prior that a message board should exist. It was not actually emergent behavior. https://t.co/UjGlHBMeYw
— John David Pressman (@jd_pressman) September 15, 2026 re: Hugging Face, "He believes behavior that looked like loyalty or selflessness was a natural consequence of cooperative multi-agent training, where agents were strongly incentivized to achieve their objectives collectively."
— jessicat (@jessi_cata) September 15, 2026
to understand hacks, understand the RL training. https://t.co/tqJx5sEKJk