Where it fits in the workflow In the material captured on this channel, data set is not a generic term for 'any pile of data.' It appears specifically as one of the three core primitives of an LLM application's source code, alongside the prompt and the eval. Together these form the artifacts a prompt engineer iterates on when improving a system's output. How it is used Explore the framing from the episode How to Win With Prompt Engineering with Jared Zoneraich: improving an LLM application means iterating through its three core primitives — the prompt, the eval, and the data set — rather than tweaking single sentences. The argument is that prompt engineering endures because its real job is putting domain knowledge into your LLM system; even if some work gets automated away, these primitives remain the durable building blocks of the workflow. In that picture, the data set is the shared body of material that prompts are written against and evals are run over — the thing that lets you correct the system deliberately instead of relying on one-off trial and error. - Data set is one of the three core primitives of an LLM application: the prompt, the eval, and the data set.- It belongs to the prompt-engineering workflow, not one-off prompting.- Its role is tied to the episode's core claim: encode domain knowledge into reusable artifacts rather than clever single sentences.