Learn
Home Foundations Glossary Research
Do
Prompts Workflows Tasks
Adapt
Domains Settings Patterns
Verify
Antipatterns Case Studies Policies Resources

Temperature

Clinical Definition

A setting in AI tools that controls how predictable or creative the output is. Lower temperature (0.0-0.3) produces more consistent, focused output, better for clinical documentation. Higher temperature (0.7-1.0) produces more varied, creative output, potentially useful for brainstorming therapy materials, but riskier for accuracy.

Technical Definition

A parameter in language model inference that scales the probability distribution over the vocabulary before sampling. Temperature = 0 selects the highest-probability token (greedy decoding); higher values flatten the distribution, increasing randomness and diversity in output.

Also known as: creativity setting, randomness parameter

Why SLPs Need to Know This

If you’re using a tool like ChatGPT, Claude, or Gemini, temperature may be adjustable in settings. Choosing the wrong temperature for your task can mean the difference between a reliable clinical document and a creative but unreliable one.

Practical Guide for SLPs

TemperatureBest ForAvoid For
Low (0.0–0.3)Progress notes, eval reports, goal writing, compliance documentationBrainstorming, generating novel therapy ideas
Medium (0.4–0.6)Parent communication, education handouts, therapy activity descriptionsLegal documents, standardized report sections
High (0.7–1.0)Brainstorming therapy activities, generating story starters, creative materialsAny clinical documentation, anything going into a record

The Clinical Analogy

Think of temperature like the difference between a structured assessment and a dynamic assessment. Low temperature is the standardized test: predictable, reliable, narrow. High temperature is the dynamic assessment: flexible, exploratory, but harder to replicate. You wouldn’t use a dynamic assessment format for an eligibility report, and you shouldn’t use high temperature for clinical documentation.

  • Hallucination: high temperature increases hallucination risk because the model is more likely to select low-probability tokens
  • Top-p / Nucleus Sampling: another parameter that controls output diversity, often used alongside temperature

SLP/IO Assistant

Powered by Claude · No PHI accepted
AI assistant for clinical workflow support. Never enter student names, DOBs, or identifiable information.
Hi! I'm the SLP/IO assistant, an opinionated AI grounded in clinical practice. I can help with goal wording, note structure, ethical reflection, and navigating LLMs responsibly. What are you working on?