looking for pretrained language models that i can readily test a draft world-state change benchmark on.
GPT is scoring essentially zero.
what’s out there that is better that I can try?
@ylecun @AntoineBordes @jaseweston still would like to test your recurrent entity networks
-
My point is that you seem to expect ML to do things it’s not designed to do.
in a certain sense that’s my point, too. but human children do it all the time
-
-
No one claims gpt*/bert/ulmfit/etc are near childlike intelligence, just that they are much better than the previous generation at nlp tasks. One step in a long chain of improvements. These models lack explicit mechanisms for logic so we don't look to them to be good at that.
-
The fact they at least sometimes make statements that are logically correctly, and sometimes handle zero-shot tasks, is just icing on the cake. The fair comparison is to what existed before, not the human brain imo.
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.