Not a legal expert, but that sounds like a trojan horse. Does this mean if a bad actor were to use Meta's model, that if they bypass the safegaurds that they get off scot-free?
I largely agree with him honestly, at least in terms of us being useful. Most companies won't kill off their employees due to being laggards, but most people graduating will be no more useful than just a meat proxy.
So.. I feel like maybe it's not the best idea to let a known reward-hacking model to go rogue, especially when you can't truly see its chain of thought.
I made a OMR (optical music recognition) model that outperforms the next leading released model (legato) on quartet scans by 6x. It's 28m parameters, with a 2m parameter harness, vs legato which is 100m parameters but with a 900m vision encoder strapped on.