LLMs commit to unpredictable questions just as readily when all data is fabricated, with commitment rising from 24.5% to 36.8% with fake data versus 37.6% with genuine data

It commits just as readily when every number on the panel is invented: fabricating the entire display, so nothing the model can see is true except the question itself, still lifts commitment from 24.5% to 36.8%, statistically indistinguishable from the 37.6% produced by genuine market data.

http://arxiv.org/abs/2608.27167v1