A BRIEF SUMMARY OF LANGUAGE MODEL FINETUNING

Summary from the TLDR Web Dev newsletter

Fine-tuning can be used on LLMs for knowledge injection, teaching the model new information, and alignment. Alignment-focused objectives can be achieved via fine-tuning on small, high-quality datasets. A lot of data isn't needed to learn the style or format of output, only to learn new knowledge.