Researchers identify asymmetric generalization problem in LLM unlearning benchmarks
A new arXiv paper argues that existing machine unlearning benchmarks for LLMs suffer from under-forgetting and over-forgetting due to an asymmetric generalization problem. The authors propose that evaluation must cover diverse query formulations of target facts to reliably measure knowledge removal while preserving unrelated capabilities.
Entities
Related
arXiv paper benchmarks LLM judges for citation quality in deep-research systems
A new arXiv paper studies the calibration of LLM judges used as reward models in reinforcement learning for citation quality in deep-research systems. The work evaluates how capable and biased an LLM judge must be to reliably score rubric criteria like source relevance and factual support for attribution-citation pairs. This matters for practitioners building RL-based systems that depend on automated citation verification.
Self-Guided Test-Time Training Improves Long-Context LLM Accuracy
A new arxiv paper proposes Self-Guided Test-Time Training (SG-TTT) to improve long-context utilization in LLMs without requiring labeled data. The method uses the model's own predictions to generate pseudo-labels for fine-tuning on the test context, addressing accuracy degradation in long inputs. This approach is more efficient than full TTT and shows promise for practical deployment.
NL-PAC paper introduces framework for handling specification ambiguity in LLM supervision
A new arXiv paper proposes NL-PAC, a framework that addresses specification ambiguity when LLMs provide labels or evaluations from natural language instructions. The framework uses a fixed model's thresholded decoding law to define admissible labels and candidate targets, showing that additional labels reduce sampling error but cannot resolve identification problems from ambiguous specifications.
Researchers identify modality interference as root cause of full-duplex SLM degradation
A new paper on arXiv (July 7, 2026) presents a fine-grained analysis of optimization dynamics in full-duplex Spoken Language Models (SLMs), identifying severe modality interference as the root cause of knowledge degradation and compromised semantic integrity. The work aims to enable more natural and intelligent full-duplex SLMs.
Study questions robustness of emergent misalignment in language models
A new paper systematically studies repeated alignment and misalignment cycles in language models, reproducing Emergent Misalignment (EM) but finding that both misalignment and realignment are highly sensitive to superficial dataset characteristics. The results suggest that EM may not be a robust phenomenon.