Developer identifies recurring patterns in LLM outputs—and why they matter

A developer catalogued distinctive 'smells'—recurring behavioral patterns and failure modes—in large language models. For indie makers building with LLMs, recognizing these patterns helps diagnose when a model is hallucinating, overthinking, or defaulting to templated responses rather than solving actual problems.