3 ms·
Personally I draw the line at the Winograd Schema Challenge. When that's passed I will be truly convinced.
by c7DJTLrn 5y ago
Personally I draw the line at the Winograd Schema Challenge. When that's passed I will be truly convinced.
- qayxc 5y agoAccording to Wikipedia, GPT-3 is almost there (88% vs ~92%) and a finetuned BERT even achieves >90%. So it stands to reason that a finetuned GPT-x would pass the test with flying colours albeit arguably not being very capable at "common sense reasoning" (whatever that should be, as "common sense" is often enough wrong). I am sceptical of such tests, as all they do is provide yet another way to game the system instead of demonstrating whatever arbitrary construct scientists use to weasel around the fact that they don't actually know what they even mean by "intelligence". IMHO we'd be better off defining characteristics for various levels/types of intelligence and trying to categorise our machines and fellow creatures accordingly.