Skip to content
From context to the next token
Separate the model’s probabilities from the rule that selects an output.
Teaching simulation · fictional values
What does temperature change?
The sky is…
Reference · T = 1
blue 60 %
grey 30 %
clear 10 %
0 %25 %50 %75 %100 %
Your setting · T = 1
blue 60 %
grey 30 %
clear 10 %
0 %25 %50 %75 %100 %
T = 1: the reference distribution is unchanged.
Only three possibilities. No real model is queried; one sample does not measure answer quality.
Check my understanding
Before answering, form your own explanation.
Increasing temperature can make an initially rare token more likely.
Takeaway
Probability expresses model preference in this context, not the truth of a sentence. Temperature changes sampling without adding knowledge. Our example has only three fictional possibilities.
Go deeperFrom concept to practicePut it into practice · ChatGPT
Scientific sources
- Vaswani et al. · 2017Attention Is All You NeedarXiv v7 · 2023Original publication
Teaching synthesis of the cited sources. Published results remain tied to their tasks, protocols and budgets.