Researcher probes what "User" means inside models, laments public indifference

A researcher explored how "User" is represented inside language models, finding it is not a simple feature, independent entity, or token span. They lamented that after months of mechanistic interpretability work, few people actually care how models work.

2026-08-31 ~ 2026-08-31 · 3 related posts

1 near-duplicate retellings: voooooogel