Why AI Lies Without Knowing It’s Lying | Victor Amit

Automated excerpt

Kadavath et al. (2022) found that models could, under the right prompting, estimate whether they could answer a question correctly, with reasonable calibration on the formats they tested. Turpin et al. (2023) showed models can give plausible written reasoning that omits the actual factor influencing the answer. The post‑trained model was less calibrated.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.