Martin Randall comments on Inference-Time Intervention: Eliciting Truthful Answers from a Language Model

[Martin Randall]

Seems like at least positive evidence for this market: