R0048 — Corporate AI Training¶
Description¶
Investigation into what standard corporate and government AI training programs teach employees about AI limitations, with particular focus on three areas: (1) general coverage of AI reliability and accuracy limitations, (2) specific warnings about sycophancy and related concepts (automation bias, overtrust), and (3) how hallucination is characterized and whether training connects it to sycophancy.
Runs¶
| Run | Date | Mode | Queries | Key Finding |
|---|---|---|---|---|
| 2026-03-29 | 2026-03-29 | Query | 3 | Training is widespread but superficial; sycophancy completely absent; hallucination-sycophancy connection not made |
| 2026-04-01 | 2026-04-01 | Query | 3 | Training widespread but shallow on limitations; sycophancy completely absent from all training; hallucination named but mischaracterized as random error |