Prompts 101

The consistency at which you can achieve quality output from an LLM is dramatically improved by having a basic understanding of how LLMs work and their core limitations.

Outputs generated by LLMs are conditional on the prompt you provide. This enables a vehicle for you to steer the generation by influencing what is the most probable output. In production systems, you will typically convert these prompts into templates, supporting a static core that has dynamic content injected such as user_details to offer tailored results.

LLMs are constrained by their context windows. This refers to the amount of data that can be attended to during generation. This is a hard constraint. If your data exceeds this limit, you must deploy a mitigating strategy, such as compaction or a sliding window.

In the below slides and video walkthrough, we provide an introductory description of the role of prompts in conditioning output and the implications of context windows.