Paper-Conference

Discursive Circuits: How Do Language Models Understand Discourse Relations?
We introduce a new task, Completion under Discourse Relation (CuDR), to reveal the underlying transformer circuits responsible for discourse understanding.
PDF Cite Code Project
Discursive Circuits: How Do Language Models Understand Discourse Relations?
Beyond In-Context Learning: Aligning Long-form Generation of Large Language Models via Task-Inherent Attribute Guidelines
LongGuide generates task-inherent metric and output-constraint guidelines to improve long-form LLM generation beyond demonstrations alone.
PDF Cite
Beyond In-Context Learning: Aligning Long-form Generation of Large Language Models via Task-Inherent Attribute Guidelines
What Makes a Good Natural Language Prompt?
A meta-analysis of prompting research that organizes natural-language prompt quality into 21 properties across six dimensions.
PDF Cite
What Makes a Good Natural Language Prompt?
LLMs Are Biased Towards Output Formats! Systematically Evaluating and Mitigating Output Format Bias of LLMs
A systematic evaluation of output-format bias in LLMs, with prompting and fine-tuning strategies for reducing format sensitivity.
PDF Cite
LLMs Are Biased Towards Output Formats! Systematically Evaluating and Mitigating Output Format Bias of LLMs
Aligning Large Language Models with Human Opinions through Persona Selection and Value-Belief-Norm Reasoning
Chain-of-Opinion selects and reasons over explicit and implicit personae to improve LLM opinion prediction and alignment.
Cite URL
Aligning Large Language Models with Human Opinions through Persona Selection and Value-Belief-Norm Reasoning
DnA-Eval: Enhancing Large Language Model Evaluation through Decomposition and Aggregation
DnA-Eval decomposes and aggregates LLM-as-judge evaluation to improve interpretability and agreement with human preferences.
Cite URL
DnA-Eval: Enhancing Large Language Model Evaluation through Decomposition and Aggregation
Monte Carlo Tree Search Boosts Reasoning via Iterative Preference Learning
A NeurIPS 2024 workshop paper on boosting reasoning with MCTS-driven iterative preference learning.
PDF Cite Code OpenReview arXiv
Multi-expert Prompting Improves Reliability, Safety and Usefulness of Large Language Models
An EMNLP 2024 prompting method that simulates and aggregates multiple expert responses to improve LLM reliability, safety, and usefulness.
PDF Cite
Multi-expert Prompting Improves Reliability, Safety and Usefulness of Large Language Models
Advancing Adversarial Suffix Transfer Learning on Aligned Large Language Models
An EMNLP 2024 study of adversarial suffix transfer learning on aligned large language models.
PDF Cite Code DOI ACL Anthology
DataTales: A Benchmark for Real-World Intelligent Data Narration
An EMNLP 2024 benchmark of 4.9k financial reports and market data for evaluating data narration by language models.
Cite DOI URL
DataTales: A Benchmark for Real-World Intelligent Data Narration