Raghuram Shankar
← Language Modelling

Part 9 of 9

The Papers & Sources

Everything cited, why it mattered, and how it was checked.

Survey methodology. Papers were verified in three ways, recorded per entry in the build: a direct arXiv Atom API round-trip; fetching the paper’s own arxiv.org abstract page and reading the title, first author and submission date off it; or verification by a survey subagent whose method is recorded in topics/language-modelling/survey/. The API rate-limited this session heavily, which is why the second route carries much of the load. No identifier on this page was written from memory — build_spec.py renders citations from survey/verified.json and marks anything unconfirmed as unverified rather than guessing. Access date: 2026-08-04. The maths follows the notation of the original papers where possible. Errors of interpretation are this page’s, not theirs.

  1. The foundations
  2. The modern block
  3. Scale, systems, and alignment
  4. Three live threads (2024 → mid-2026)
  5. The complete reference list
  6. Suggested reading order
  7. Quiz: 5 questions