Temperature is not a remedy A reflexive objection from practitioners familiar with LLM configuration holds that increasing sampling temperature would attenuate these distributional biases by flattening the probability landscape from which characters are drawn. Irregular’s empirical results are unambiguous in… Read More "LLM-generated passwords are indefensible. Your codebase may already prove it"