DM simulates a lower fidelity version of real world physics → Applies real world AI methods → Achieves generalised AI performance.
This is a pretty concrete demonstration that current AI methods are sufficient to achieve generality, just need more real world data to match the more complex physics of reality.
This seems to support Reward is Enough.
More specifically:
DM simulates a lower fidelity version of real world physics → Applies real world AI methods → Achieves generalised AI performance.
This is a pretty concrete demonstration that current AI methods are sufficient to achieve generality, just need more real world data to match the more complex physics of reality.
And more compute of course. As always.
[Deleted]