Fine-tuning constraints model behavior

Adjusting the weights of a model directly aligns it to specific tasks or tones more permanently than prompt engineering. It is infrastructure-level configuration, much like how Infrastructure as code ensures reproducible environments.¹

Footnotes

1. Ouyang, L., et al. (2022). Training language models to follow instructions with human feedback. NeurIPS. ↩