Reference library
Lost in the Middle: How Language Models Use Long Contexts
Paper summary
Controlled question-answering and retrieval tasks show that relevant information is used unevenly across long contexts. Performance often falls when the needed material occurs in the middle rather than near the beginning or end.
Relevance to the project
Useful for testing whether F actually uses its bounded histories and slots. Vary relevant signal position instead of equating available capacity with functional access.
Text-context effects need not transfer unchanged to local recurrent mechanisms.
Source record
- Reference ID
- lost-in-middle
- Archived version
- 2307.03172v3
- Summary provenance
- reference-library/papers/lost-in-middle/metadata.json · summary
- Review status
- initial orientation summary; not a critical full-paper review
- Reading scope
- Selected abstract/opening and available closing discussion/conclusion passages in the preserved text extract; broader project notes provide context. No reproduction of experiments, exhaustive methods review, or equation/figure audit.