LLM
2 essays on this topic.
Treat the LLM Like an Unreliable API: Designing AI Pipelines That Hold Up
An LLM call is a slow, costly, non-deterministic dependency. Design for it like any flaky API: validate, retry, log, evaluate and keep a human in charge.
Generative AI Large Language Model – Train From Pre-trained Model
What is a Large Language Model, how does it work, and can you train your own? A plain-English guide, from tokenization to fine-tuning a pre-trained GPT-2.
