Publications

Refusal tokens enable controlling a single model's refusal rates without the need of any further fine-tuning. (COLM)

An agentic exploration framework inspired by human curiosity. (CoRL)

An open-source initiative dedicated to developing high-quality, culturally relevant data for SEA languages. (ACL)

The ability of LMs to process Javanese honorifics through classification and machine translation tasks. (ACL)

Exploring knowledge loss in multilingual LMs, focusing on linguistic differences affecting representational learning. (ACL)

Loss deceleration and ZSL provide new insights into the training dynamics underlying language model scaling laws. (ACL)

A critical re-examination of popular out-of-distribution (OOD) detection procedures. (ICML)

Specialized classifiers that evaluate text based on predefined trustworthiness objectives. (ICML)

Introducing APT, an adversarially pre-trained transformer achieving SOTA on small tabular tasks. (ICML)