Deterministic vs Stochastic
A deterministic process always produces the same output for the same input, while a stochastic process involves randomness and may produce different outputs each time.
In AI
Model training is stochastic (random initialization, data shuffling, dropout). Inference can be deterministic (temperature=0, greedy decoding) or stochastic (sampling with temperature>0). Setting random seeds enables reproducibility.
Practical Impact
Stochasticity in training means different runs produce different models. In deployment, stochastic generation (sampling) produces diverse outputs but makes exact reproduction impossible. Deterministic settings are preferred for testing and debugging.