Xiang Lisa Li (@xianglisali2) 's Twitter Profile
Xiang Lisa Li

@xianglisali2

PhD student at Stanford

ID: 1134226884818919425

calendar_today30-05-2019 22:35:37

33 Tweet

2,2K Followers

227 Following

Xiang Lisa Li (@xianglisali2) 's Twitter Profile Photo
Xiang Lisa Li
@xianglisali2

4 months ago

arxiv.org/abs/2407.08351 LM performance on existing benchmarks is highly correlated. How do we build novel benchmarks that reveal previously unknown trends? We propose AutoBencher: it casts benchmark creation as an optimization problem with a novelty term in the objective.

arxiv.org/abs/2407.08351 LM performance on existing benchmarks is highly correlated. How do we build novel benchmarks that reveal previously unknown trends? We propose AutoBencher: it casts benchmark creation as an optimization problem with a novelty term in the objective.