Breaking Down AI's Diversity Dilemma in Subjective NLP
Ensemble Diversity Optimization (EDO) tackles the tricky subject of annotator disagreement in NLP, offering a fresh approach to model uncertainty. But will it actually change the game?
Read the full articleYou might also wanna read
DecompR: A Method for Reducing Weighting Noise in Multi-Stakeholder LLM Alignment
Multi-stakeholder tasks require one output to satisfy users with conflicting preferences. Holistic LLM judges conflate utility estimation an
arxiv.org·1mo ago
Verbalized Sampling: A Training-Free Method to Mitigate Mode Collapse and Improve LLM Output Diversity
Post-training alignment often reduces LLM diversity, leading to a phenomenon known as mode collapse. Unlike prior work that attributes this
arxiv.org·14d agoVerbalized Sampling: A Training-Free Method to Mitigate Mode Collapse and Improve LLM Output Diversity
Post-training alignment often reduces LLM diversity, leading to a phenomenon known as mode collapse. Unlike prior work that attributes this
arxiv.org·14d ago
EntropyLong: Using Predictive Uncertainty to Improve Long-Context Language Model Training
Training long-context language models to capture long-range dependencies requires specialized data construction. Current approaches, such as
arxiv.org·8mo ago
[ICML 2026] The Flexibility Trap: Rethinking the Value of Arbitrary Order in Diffusion Language Models
Authors: Zanlin Ni, Shenzhi Wang, Yang Yue, Tianyu Yu, Weilin Zhao, Yeguo Hua, Tianyi Chen, Jun Song, Cheng Yu, Bo Zheng, Gao Huang
The Significance of Generalization in AI Systems and the Quest for Consciousness
This blog post outlines a key engineering principle I’ve come to believe strongly in for building general AI systems with deep learning. Thi
Bidirectional Evolutionary Search: A New Framework for Self-Improving Language Models
Search has been proposed as an effective method for self-improving language models and agentic systems, both for post-training sample genera
arxiv.org·1mo ago
Comments
Sign in to join the conversation.
No comments yet. Be the first.