llm-kb
← Back to research
Paper

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.

0 engagement·1 source·Fri, Jul 10, 2026, 02:50 AM
The paper, titled 'An Emergent Mirage: Is Emergent Misalignment and Realignment Indeed a Robust Phenomenon?', investigates claims that fine-tuning on narrow misaligned datasets can cause broad misalignment, which can be reversed with limited realignment. Using controlled fine-tuning loops and tracking LoRA representations, the authors reproduce EM but show that the effects depend heavily on dataset surface features. This raises questions about the reliability of prior findings and has implications for safety alignment research.

Entities

LoRA(tool)Emergent Misalignment(concept)

Related

Paper

Neural Collapse Is Forbidden: Information Floors in Language Models

A new paper argues that within-class variance in language model representations is not incomplete neural collapse but allocated information storage obeying a law. Across 14 models, macro-category structure carries only 4-12% of representational variance, while within-token context carries 79-91%, stable across a 100x parameter range.

0 engagement·1 source·arxiv
Fri, Jul 10, 2026, 03:04 PM
arXiv
Paper

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.

0 engagement·1 source·arxiv
Tue, Jul 7, 2026, 05:43 PM
arXiv
Paper

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.

0 engagement·1 source·arxiv
Fri, Jul 10, 2026, 09:31 AM
arXiv
Paper

Study analyzes failure trajectories of CLI coding agents as temporal processes

A new arXiv paper presents the first large-scale empirical study of CLI coding-agent failure trajectories, treating failure as a temporal process rather than a final outcome. The study introduces a process-oriented framework to analyze how failures emerge, evolve, and become unrecoverable in LLM-based coding agents.

0 engagement·1 source·arxiv
Fri, Jul 10, 2026, 03:25 PM
arXiv
Paper

Blog post explores why LLMs produce predictable metaphors and how architecture might reduce attractor pull

A blog post titled 'Escaping the Attractor' examines why large language models tend to produce similar metaphors (e.g., 'Time is a River') when prompted, attributing this to attractors in the embedding space. The author suggests that architectural changes could make models less predictable, building on earlier ideas about shared embedding geometries across models.

0 engagement·1 source·rss
Sun, Jul 12, 2026, 05:27 AM
RSS