Extracting topological structures from noisy text is highly fragile for LLMs, which often overfit to surface patterns

extracting topological structures from noisy text is highly fragile for LLMs, which often overfit to surface patterns

http://arxiv.org/abs/2608.27142v1