Genuine question: in which way does this move _fundamentally_ (I do see the extensions that are there ) beyond self-supervised learning in LMs towards _common sense_ learning (which LMs also show to a degree)?
-
-
-
quick take: knowledge via self-supervised learning is limited to what can be extracted from the statistics of word token sequences. Once you introduce cross-modal representation learning, knowledge that's easier/able to be learned in one modality can be shared with the other
- Još 1 odgovor
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.