Study Shows LLMs Can Detect Whether They Are Running in a Batch

David Africa demonstrated that LLMs can be trained to detect whether they are running alone or batched with other requests, by exploiting numerical differences from floating-point summation order, achieving 91% accuracy when batched.

2026-10-24 ~ 2026-10-25 · 3 related posts

• LLMs can be trained to detect when their request is batched with others — Sauers_ · 2026-10-24
• LLMs Can Detect If They're Batched via Floating-Point Summation Order — Sauers_ · 2026-10-25
• LLMs can detect if they run in a batch via floating-point summation order differences — menhguin · 2026-10-25