# Temperature **Entity class:** Concept **Research spine:** [[research/Chronology of Signals, Silos, and Sovereign Software - Research|Chronology of Signals, Silos, and Sovereign Software]] ## Definition Temperature is a sampling control that reshapes a language model's probability distribution before the next token is selected. ## Chronological Proximity Its importance became visible in 2019 discussions of neural text degeneration and later demonstrations of GPT-2 generation. ## Relational Proximity Lower temperature concentrates probability and can produce repetition; higher temperature spreads probability and can produce incoherence; it operates after [[wiki/Next-Token Prediction|next-token probabilities]] are computed. ## Collection Routes [[collections/The Civilian Intelligence Marketplace|The Civilian Intelligence Marketplace]] · [[collections/Terrorism, Counterterrorism, and the Intelligence Environment|Terrorism, Counterterrorism, and the Intelligence Environment]] · [[wiki/Signals, Silos, and Sovereign Software|Signals, Silos, and Sovereign Software]] ## Sources / Provenance - [[research/Chronology of Signals, Silos, and Sovereign Software - Research|Chronology of Signals, Silos, and Sovereign Software - Research]].