11 papers
From Interpretability to Control: Insights from Six Years of the TrustNLP Workshop
Rahul Gupta, Abhinav Mohanty, Anaelia Ovalle +10
The Workshop on Trustworthy Natural Language Processing (TrustNLP), co-located with major ACL conferences since 2021, has grown from 8 proceedings papers to 41 over six editions, d…
Emergent Strategic Reasoning Risks in AI: A Taxonomy-Driven Evaluation Framework
Tharindu Kumarage, Lisa Bauer, Yao Ma +7
As reasoning capacity and deployment scope grow in tandem, large language models (LLMs) gain the capacity to engage in behaviors that serve their own objectives, a class of risks w…
Sparrow: Sparse Rollout for Stable and Efficient Long-context RL of Large Language Models
Yang Zhou, Ranajoy Sadhukhan, Zhaofeng Sun +7
Despite being powerful, reinforcement learning with verifiable rewards (RLVR) induces extremely long COT, making it computationally expensive. Since RLVR per-step cost is dominated…
IdleSpec: Exploiting Idle Time via Speculative Planning for LLM Agents
Daewon Choi, Kyunghyun Park, Woomin Song +4
Large language model (LLM)-based agents solve complex tasks by leveraging multi-step reasoning with iterative tool calls and environment interactions, which incur idle time while w…
ExComm: Exploration-Stage Communication for Error-Resilient Agentic Test-Time Scaling
Woomin Song, Beomjun Kim, Daewon Choi +4
A common failure mode in long-horizon agentic test-time scaling is error propagation, where factual errors or invalid deductions introduced at intermediate steps persist in the age…
RoPE Distinguishes Neither Positions Nor Tokens in Long Contexts, Provably
Yufeng Du, Phillip Harris, Minyang Tian +5
We identify intrinsic limitations of Rotary Positional Embeddings (RoPE) in Transformer-based long-context language models. Our theoretical analysis abstracts away from the specifi…