Context rot
Also: long-context degradation
A drop in answer quality as the context grows or fills with clutter. It does not happen in the same way in every model or task.
The fuller definition
Context rot is the informal name for declining output quality as a conversation or task context grows long. Earlier instructions get dropped and corrected facts creep back in.
2,000
8,000
4,000
2,000
A hypothetical 16,000-token budget. Real limits and token accounting depend on the model. Saving a file outside this budget does not mean it is present in the current context, and a passage inside it is not guaranteed equal attention.
A concrete example
After forty exchanges, an assistant stops following a formatting rule given at the start and reintroduces a date you corrected.
The distinction that matters
Fitting inside the context window does not mean every passage is used equally. Start a fresh session or compact the history when quality slips.
Related terms
Context window
How much text the model can hold in mind at once: your question, the documents you added, and the conversation so far. Anything outside it is invisible to the model.
Prompting & contextCompaction
Squeezing a long conversation into a short summary so there is room to keep going. Details can be lost in the squeeze.
Prompting & contextMemory
A feature that saves facts between conversations so you do not have to repeat them. It is a saved note, not the model learning.
Where it is taught
Does AI learn from your corrections?Source and further reading
Anthropic · Context engineeringThe definition and example are written for this library. The link gives background, official documentation or legal authority as labelled. Legal scope depends on the jurisdiction and current text; product features depend on release and settings.