If you want to use external data, or hard-code Core Knowledge priors -- all of this is allowed. Virtually anything is allowed. Just get those leaderboard scores to go down :)
Show this thread
If you're not sure where to start on the ARC competition, here's my advice. It's a very difficult challenge, but I strongly believe (with some evidence) that someone smart & motivated can develop in a few weeks an approach that solves ~5-10% of the tasks in the hidden test set.pic.twitter.com/1hvZrXugnt
-
-
-
Also, if you're curious why ARC looks the way it does, where it comes from, and what it means to work on solving it, you should check out this paper:https://arxiv.org/abs/1911.01547
Show this thread
End of conversation
New conversation -
-
-
Galaxy-brain solution: train a text-generating AI to generate code to manually solve the problem
-
I think DL text generation has *some* potential to be helpful in code generation, but very little. What it does is interpolation, pattern recognition -- but programming is ~90% abstraction and reasoning.
- Show replies
New conversation -
-
-
Francois is of course exaggerating that you won't find any useful recent literature. I would recommend reading our recent paper that deals with exactly this kind of problem. Here is a snippet from the discussion section. https://robotics.sciencemag.org/content/4/26/eaav3150.full?ijkey=9p/p9D23WW2Ek&keytype=ref&siteid=robotics …pic.twitter.com/B1IcRU13Y2
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
This is such a motivating challenge - no established path!
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.