paper

Do LLM Embedding Spaces Recover Expert Structure?

paperactiveprovisionaldo-llm-embedding-spaces-recover-expert-structure--d560b099·1 events·first seen 36h ago

Aliases: Do LLM Embedding Spaces Recover Expert Structure?

Co-occurring entities

More like this (12)

LCO-Embedding A sleep-like consolidation mechanism for LLMs Continual LLM Upcycling: A Predictor-Gated Bank-Wise Sparsity Training Recipe for Dense-to-Sparse LLMs Tying the Loop -- Tied Expert Layers in Mixture-of-Experts Language Models Sparse Embedding Models LLM-based content classification The Masked Advantage: Uncovering Local-Language Access to Cultural Knowledge in LLMs LLM inference Instructional Segment Embeddings Which Models Are Our Models Built On? Auditing Invisible Dependencies in Modern LLMs Backdoor Unlearning Generalization: A Path Toward the Removal of Unknown Triggers in LLMs long-context LLMs

Recent events (1)

4arXiv · cs.CL·36h ago·source ↗

LLM embedding spaces partially recover expert-defined symptom structure in mental health language

A new arXiv preprint investigates whether LLM embedding geometry aligns with expert-defined symptom structure in mental health language, using 28 Reddit communities as a testbed. The authors compare pretrained and fine-tuned Qwen3 embeddings (0.6B and 4B) against an expert symptom matrix via representational similarity analysis, with controls for affective, stylistic, and topic confounds. Results show measurable but level-dependent alignment: fine-tuning strengthens it at fine-grained category levels, and larger scale improves both zero-shot alignment and fine-tuning gains. The paper argues that classification accuracy alone is insufficient to validate embedding geometry against domain knowledge.

Evaluation and Benchmarking Reddit Do LLM Embedding Spaces Recover Expert Structure?Qwen3