IMO, using the API rather than pixels doesn't make the results any less impressive. Vision is a much more mature problem than the challenges in strategy and long-term planning, which are the core contribution of OpenAI's work.
Re @OpenAI and DotA & recent debates, there are 2 distinct issues: is result impressive (yes) and is it playing same game as humans (no, see eg thread below).
Kudos to @OpenAI blog for noting use of API rather than pixels; many of us now trying to understand implications.https://twitter.com/chavezaaronj/status/1031563766129811456 …
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
'playing the same game' is a red herring as Dijkstra pointed out in a different context. If anything, the visual representation has a much better saliency than a raw log of all the information with no hints as to what matters.
-
Rather than focus on the limits, I'd rather talk about a) the differences b) the technological infrastructure (how did this piece get glossed over so completely when it's one of the crowning achievements of their work) c) the relatively 'vanilla' NN architecture used
- 7 more replies
New conversation -
-
-
You’re right, they aren’t playing the same game. Humans can communicate through language, and have an ability to one shot learn
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
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.