takedown of OpenAI’s apparent success at ‘Dota 2’https://apple.news/A45CaowNkR7GQ8x92qTIbdg …
-
It's pure self-play, so no labeled databases, but I think there is real disagreement about how central the role of input representation is to the overall problem. Is learning a complex policy from hand-crafted representations solving 90% of the game, or 10%?
I was responding to @smerity’s specific suggestion above of using a traditional pipeline w human labeling to do the vision work that OpenAI abstracted away from
-
-
I actually disagree human annotation is needed - the API already provides that, after all. Learning vision in sim w full state info is pretty important to robotics nowdays, eg http://carla.org . I also dont think claim here is humanlike learning, just humanlike performance
-
Regarding human annotation I was saying that I believe a standard vision pipeline with human annotated events would work if we didn't have an API at all (i.e. if it was a blackbox and unmodifiable game / real life camera feed) and enable all the work that
@OpenAI Five has on top.
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.