Addendum 2 cont'd
* prompting (prompt engineering) used to guid LM responses consistent w. desired outcome
* chaining decomposes complex tasks into smaller, manageable components
* prompt chaining:
1. create of concise summary of orig. document
2. semantic search for related texts, annotations f. training corpus
3. prompt for task-based label assignment
* prompt chaining:
1. enhances performance over zero-shot
2. surpasses micro-F1 score of larger models (ChatGPT zero-shot) c. smaller models