This all-or-nothing approach to progress is pretty silly. I'm sure it's a toy model, and I'm sure you'll find flaws with it, but that doesn't mean it's not progress.
. @Ylecun are you arguing that your team has a robust solution to the problem of getting deep nets to understand the causal consequences of events as they unfold over time, or just pointing me to a toy model? Have you tried it on the examples in this thread?https://twitter.com/ylecun/status/1188902027495006208 …
-
-
-
Hey, I'm all for taking steps, but Dr
@ylecun told me the problem was solved, period, and that's completely ridiculous. - 1 more reply
New conversation -
-
-
Typical fallacious rhetoric, "feature A works in a toy model X", "feature B works in a toy model Y" hence (falsely) conclude - "features A and B must work in production code C". I've seen this a lot, not only from this particular specimen.
-
It's almost not worth doing. Many researchers (e.g. Hal Daume) are supercritical of such toy tasks and babi in particular. https://twitter.com/haldaume3/status/669883787967262720 … The task is super-contrived, limited grammar & vocabulary. I bet you could defeat it by use of conjunctions & a thesauruspic.twitter.com/6viPhGm8cx
End of conversation
New conversation -
-
-
Closest I've seen recently is competitors in the
@allen_ai Allen Institute's ARC multichoice challenge https://leaderboard.allenai.org/arc/submissions/public … leader is https://arxiv.org/abs/1909.11764 that uses a corpus of 14m docsThanks. 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.