How Language Models Fail: Token-Level Signatures of Committed and Persistent Reasoning Failures
Characterizing language model reasoning failures through token-level uncertainty signals, and showing they arise through two empirically distinguishable processes: committed failure, where a model locks onto a wrong path early, and persistent uncertainty, where uncertainty accumulates throughout the trace.
Failures in language model reasoning emerge through distinct processes that leave identifiable signatures in the reasoning trace. We characterize these failures using token-level uncertainty signals, finding they arise through two empirically distinguishable processes. The first is committed failure, in which a model locks onto an incorrect reasoning path early in its trace. A central diagnostic signature is the commitment point, beyond which considering additional tokens hurt rather than help failure detection. In the second, persistent uncertainty, uncertainty instead accumulates throughout, and the full trace is needed to best distinguish failing from successful completions. These signatures reproduce across 23 model-dataset configurations, with the framework's falsifiable predictions holding in 20 of 23 cases, well above chance across both failure modes. Finally, we demonstrate our failure mode framework has direct implications for self-consistency, identifying when uncertainty signals complement it and when it can be selectively skipped. These results offer a foundation for understanding when LLM reasoning failures become detectable and for adapting detection strategies accordingly.
Joint work with Tanvi Thoria, Marc R. Schlichting and Mykel J. Kochenderfer at Stanford University.