Ike
@ikeo.bsky.social
PhD Candidate @Purdue University. Research at the intersection of Human-AI Interaction, HRI, and Computational Human Values.
Incredibly grateful 🙏 for this year. Very lucky. X2 25
January 1, 2025 at 4:57 AM
Incredibly grateful 🙏 for this year. Very lucky. X2 25
1) Information utility values (information seeking, wisdom/knowledge) were the most dominant human values in the preference examples. In contrast, prosocial values (animal rights and tolerance, etc.) were significantly underrepresented, thus showing an imbalance in values encoded into LLMs.
December 16, 2024 at 1:19 PM
1) Information utility values (information seeking, wisdom/knowledge) were the most dominant human values in the preference examples. In contrast, prosocial values (animal rights and tolerance, etc.) were significantly underrepresented, thus showing an imbalance in values encoded into LLMs.
Just wrapped up an incredible week at #NeurIPS 2024, where I presented our spotlight paper ⭐️ 🌟 😀 Value Imprint - A Technique for Auditing Human Values Embedded in RLHF Datasets. Through our work, we audited three prominent available RLHF datasets to examine values encoded in them. We found that:
December 16, 2024 at 1:19 PM
Just wrapped up an incredible week at #NeurIPS 2024, where I presented our spotlight paper ⭐️ 🌟 😀 Value Imprint - A Technique for Auditing Human Values Embedded in RLHF Datasets. Through our work, we audited three prominent available RLHF datasets to examine values encoded in them. We found that: