Recent papers show LLMs achieve near-random accuracy in causal relation classification, raising questions about whether such failures arise from limited pretraining exposure or deeper representational gaps. We investigate this under uncertainty-based evaluation, testing whether pretraining exposure to causal examples improves causal understanding using >18K PubMed sentences—half from The Pile corpus, half post-2024—across seven models (Pythia-1.4B/7B/12B, GPT-J-6B, Dolly-7B/12B, Qwen-7B). We analyze model behavior through: (i) causal classification, where the model identifies causal relationships in text, and (ii) verbatim memorization probing, where we assess whether the model prefers previously seen causal statements over their paraphrases. Models perform four-way classification (direct/conditional/correlational/no-relationship) and select between originals and their generated paraphrases. Results show almost identical accuracy on seen/unseen sentences (p>0.05), no memorization bias (24.8% original selection), output distribution over the possible options almost flat — with entropic values near the maximum (1.35/1.39), confirming random guessing. Instruction-tuned models show severe miscalibration (Qwen: >95% confidence, 32.8% accuracy, ECE=0.49). Conditional relations induce highest entropy (+11% vs direct). These findings suggest that failures in causal understanding arise from the lack of structured causal representation, rather than insufficient exposure to causal examples during pretraining.

The recognition of mental health’s crucial significance has led to a growing interest in utilizing social media text data in current research trends. However, there remains a significant gap in the study of panic and anxiety on these platforms, despite their high prevalence and severe impact. In this paper, we address this gap by presenting a dataset consisting of 1,930 user posts from Quora and Reddit specifically focusing on panic and anxiety. Through a combination of lexical analysis, emotion detection, and writer attitude assessment, we explore the unique characteristics of each condition. To gain deeper insights, we employ a mental health-specific transformer model and a large language model for qualitative analysis. Our findings not only contribute to the understanding digital discourse on anxiety and panic but also provide valuable resources for the broader research community. We make our dataset, methodologies, and code available to advance understanding and facilitate future studies.