AIThe case for zero-shot classifiers on tasks handled by LLMs
A user argues that many problems solved with LLMs could have been handled by zero-shot classifiers, and wishes those classifiers had been scaled as much as decoder-only models.
TLDR
A user makes the case that zero-shot classifiers could have solved many problems now tackled with LLMs, lamenting that they weren't scaled as much as decoder-only models. In a follow-up reply, they clarify that this isn't specifically an argument about Jev, saying “we don't know what it is.” They instead emphasize their appreciation for BERT, praising its simplicity.
Combined views
979
1 Source, first seen 5h ago
The case for zero-shot classifiers on tasks handled by LLMs
A user argues that many problems solved with LLMs could have been handled by zero-shot classifiers, and wishes those classifiers had been scaled as much as decoder-only models.
TLDR
A user makes the case that zero-shot classifiers could have solved many problems now tackled with LLMs, lamenting that they weren't scaled as much as decoder-only models. In a follow-up reply, they clarify that this isn't specifically an argument about Jev, saying “we don't know what it is.” They instead emphasize their appreciation for BERT, praising its simplicity.