Low-quality data may force models to build extra machinery, capping pretraining quality
davidmanheim · x · 2026-09-09
In a discussion with Toby Ord and Dwarkesh Patel, David Manheim argues that during pretraining, including low-quality data forces the model to construct additional complex internal machinery beyond correct capabilities — implying an effective upper limit on data quality tradeoffs. Toby Ord responds that breaking out "gains from data filtering" as a distinct category is a useful addition to his mental model.
Related event: Scholars Debate Whether Pretraining Scaling Is Slowing Down(4 posts)→
More from Research
- Token Collapse: Study Finds Graph Soft Prompting Tokens in LLMs Fail to Encode Structure — Sapienza · 2026-10-27
- Mathathon 2.0 launches with Terence Tao's SAIR, $1.5M in compute for teams — _sathvikr · 2026-10-27
- The 'sweet lesson' is real: two kinds of ML problems need opposite strategies — mike64_t · 2026-10-27
- PixGR: Pixel-space generative decoding for fast feed-forward 3D reconstruction — kwangmoo_yi · 2026-10-27
- UniAction: unified action conditioning lets one world model handle mixed action types — ZeYanjie · 2026-10-27
- Infini-AI Lab open-sources Hermit, a framework for agent harness evolution and continual learning — BeidiChen · 2026-10-27