LLMs Confidently Fabricate Meaning From Pure Noise Data
Experiments by Schema Labs engineers and Reddit users showed that when given tables of purely random numbers, all tested LLMs confidently fabricated plausible interpretations like sensor readings, revealing a lack of skepticism toward meaningless data.
2026-09-02 ~ 2026-09-02 · 2 related posts
- Feed LLMs a table of pure noise and they'll confidently invent 'sensor data' — No-Plant-5234 · 2026-09-02
- Every Model Tested Described a Table of Pure Noise Plausibly — No-Plant-5234 · 2026-09-02