The Oil Refinery Analogy for AI Context Windows: Turning Raw Data into Rocket Fuel

When presenting the value Four/Four brings to the research and discovery space, I often use an oil refinery analogy (A-level science throwback) to explain how the sausage is (and should be) made.

It’s back to the context window problem. Think of the context window as the refinery's processing unit, a fixed-capacity container that can only hold and work on so much material at once. Everything you feed into it (documents, search results, code, prior messages) is raw crude oil. It’s a mix of valuable and worthless material, all mixed together, taking up valuable space.

Each stage of the refinement process yields varying quality outputs, and btw there are lots of stages to run through! The further down the refinement process you go, the more valuable the resulting output.

This refined output is what actually goes into the context window. High-octane rocket fuel rather than crude sludge that clogs the engine (burns tokens, dilutes attention, and can even cause "knock," i.e. the model latching onto irrelevant details or getting confused by noise).

You see, we only ever feed the Four/Four rocket ship with rocket fuel for peak performance!