Prompt Engineering, RAG, and Fine-tuning: Benefits and When to Use

- Prompt engineering involves priming a prompt, handling edge cases, ignoring irrelevant information, and formatting the output.

• Retrieval augmented generation (RAG) adds dynamic content to prompts by retrieving external knowledge to enhance the model's responses.
• To implement RAG, a database with information is needed, text is split and converted into embeddings for retrieval, and the most relevant information is inserted into the prompt.
• Fine-tuning involves training a model on prompt completion pairs to teach intuition and improve output quality.
• Fine-tuning can optimize speed and cost, narrow the range of possible outputs, and bake in style, tone, and formatting to the model's responses.
• Fine-tuning can work together with RAG by training models on examples and improving response quality.
• Fine-tuning can be used to create a scalable layer of training data to improve model behavior over time.
• Fine-tuning can significantly impact response times and cost, making it a cost-effective option for large volumes of requests.
• Prompt engineering, RAG, and fine-tuning all aim to improve model outputs and can be used together as tools for working with large language models.

Researchers/sources

• Mark Hennings
• Entrypoint Ai