Why did it lose track of something I already told it?
Context window
A model works with a bounded amount of input at a time. A product may select, retrieve or summarize previous material before it reaches the model.
An everyday example
An early constraint is missing from the assistant’s later draft. You restate the constraint in a short brief and check the revised output.
Original illustration, not an observed participant case.
Where the idea stops
Fitting inside a large context window does not guarantee that every detail will be used reliably. Performance depends on the task and model.
Something to try
Create a short task brief: goal, constraints, authoritative sources, unresolved questions and acceptance checks. Keep the sources available.
Practise on a fictional case →Go deeper: the builder’s view
For builders: test information placement, truncation and summarization. The cited experiments are historical evidence, not a universal score for current models.
Source and interpretation
This is a plain-language interpretation with an applied example; it is not a quotation or an endorsement from the source.
Liu et al. — Lost in the Middle: How Language Models Use Long Contexts (2023) ↗Catalog v0.1 · checked 2026-10-29