v2 draft of related work (FROST)

danton-nlp / GEF

1 stars 0 forks source link

v2 draft of related work (FROST) #53

Closed dleve123 closed 2 years ago

dleve123 commented 2 years ago

Upon a closer read of FROST paper, it seems like FROST is not geared at specifically improving factuality, and as such does not require a direct comparison against.

[x] Mention FROST & clearly argue how it's different from our approach
[x] Merge with constrained beam search to faithfulness section

dleve123 commented 2 years ago

Addressed FROST a bit. It's short, but I think it's clear how it's faithfuless-oriented due to the fact that it extracts entities from the source (and is fundamentally intrinsic)