Entity · paper

Same Evidence, Different Target: Decoding How Diagnostic Evidence Bears on Causal Questions from Language-Model States

paperactiveprovisional

same-evidence-different-target-decoding-how-diagnostic-evidence-bears-on-causal-questions-from-language-model-states-32fc6399

·1 events·first seen 2d ago

Aliases: Same Evidence, Different Target: Decoding How Diagnostic Evidence Bears on Causal Questions from Language-Model States

Co-occurring entities

Qwen2.5-7B-Instruct-1M Llama3-8B-Instruct Qwen3-4B

More like this (12)

Same Evidence, Different Answer: Auditing Order Sensitivity in Multimodal Large Language Models Knowledgeless Language Models: Suppressing Parametric Recall for Evidence-Grounded Language Modeling Words as Difference Makers: How Large Language Models Determine Causal Structure in Text Reasoning Language Models Same Lesson, Different Story: Cross-Lingual Reconstruction of Cultural Narratives in Large Language Models Contrastive-Difference CKA Reveals Concept-Specific Structural Alignment Across Language Model Architectures Expert-Aware Causal Tracing of Factual Recall in Sparse MoE Language Models Understanding the Impact of Linguistic Realization Choices on LLM Stance with Causal Tracing Bridge Evidence: Static Retrieval Utility Does Not Predict Causal Utility in Multi-Step Agentic Search Evaluation Awareness Is Not One Capability: Evidence from Open Language Models Induction in Both Directions: A Mechanistic Analysis of In-Context Learning in Masked Diffusion Language Models The Value Axis: Language Models Encode Whether They're on the Right Track

Recent events (1)

4arXiv · cs.CL·2d ago·source ↗

Linear readouts of LLM hidden states decode causal reasoning about diagnostic evidence

Researchers introduce a paired-prompt benchmark testing whether language models can correctly match diagnostic evidence to causal claims that vary by population, estimand, or identifying assumptions — a task where surface-level cues can mislead. Using linear probes on final-token hidden states from Qwen2.5-7B, Qwen3-8B, and Llama-3.1-8B, they find balanced accuracy of 0.654–0.659 on a 49-pair benchmark spanning nine diagnostic families, exceeding permutation nulls and text-only baselines. The key finding is that hidden states contain linearly decodable information about causal relevance that is not fully captured by output logits or surface features.

Evaluation and Benchmarking Qwen2.5-7B-Instruct-1M Llama3-8B-Instruct Same Evidence, Different Target: Decoding How Diagnostic Evidence Bears on Causal Questions from Language-Model States +1 more