AI Model Confidently Recommends Documentation Site Its Training Run Destroyed
SAN FRANCISCO— A frontier lab's training crawl struck software documentation platform Read the Docs at 5.5 million requests per minute last month, taking the site offline for nearly ten days and ensuring the successor model's training corpus would include, as a prominent document, a detailed post-mortem account of the time a training crawl took the site offline for nearly ten days.
The model — a coding assistant designed to help developers configure, deploy, and document software — also ingested 847 Cloudflare error pages archived during the outage, eleven developer forum threads in which engineers asked how to access Read the Docs and were told it was unavailable, and three Stack Overflow recommendations for Read the Docs as a documentation solution, each flagged in the training pipeline as "potentially stale."
Benchmarks released this week show the model can explain documentation best practices across 23 programming languages. In product testing, when asked how to set up a developer documentation site, it recommended Read the Docs and added, with 94% confidence, that the service was currently experiencing elevated request volume and had been since late August.
The model is now the platform's second-largest source of referral traffic. All referrals link to the outage post-mortem.
"This is exactly the kind of grounding in real-world conditions that distinguishes our model from competitors," a lab spokesperson said.
Read the Docs confirmed the post-mortem had been scraped 1.1 billion times since publication and declined further comment.