Confident wrong answers may show up as instability, not doubt
The hardest kind of is an answer that sounds smooth, confident, and believable but is wrong. Methods that look for the model sounding unsure are weak in this case. Across about 124 prompts, confident made-up answers and true answers looked nearly the same when only the average internal response was compared.
The model did not show a steady “lying direction.” The clearer difference was size and variance. Made-up answers caused larger and more spread-out movement inside the model as it generated text. On the , the variance was about 7 times higher for made-up answers than for true ones, with Cohen's d around 0.58 and a p value around 0.005.
The movement grew as the level of fabrication grew, which supports the idea that this is tied to rather than random noise. The practical advice is to detect instability, measure it across the whole generated answer, and connect the detector to a follow-up action such as checking or correcting the answer instead of using it as a simple pass-or-block gate.
Key points
- Confident wrong answers are hard to catch by looking for .
- About 124 prompts showed little average internal difference between truth and confident fabrication.
- The stronger signal was variance, not a consistent lying direction.
- Made-up answers showed about 7 times more variance on the .
- A detector should trigger checking or correction, not act as a standalone pass-or-block rule.