We present {that a} GPT-3 mannequin can study to specific uncertainty about its personal solutions in pure language—with out use of mannequin logits. When given a query, the mannequin generates each a solution and a stage of confidence (e.g. “90% confidence” or “excessive confidence”). These ranges map to possibilities which can be properly calibrated. The mannequin additionally stays reasonably calibrated underneath distribution shift, and is delicate to uncertainty in its personal solutions, reasonably than imitating human examples. To our information, that is the primary time a mannequin has been proven to specific calibrated uncertainty about its personal solutions in pure language. For testing calibration, we introduce the CalibratedMath suite of duties. We evaluate the calibration of uncertainty expressed in phrases (“verbalized chance”) to uncertainty extracted from mannequin logits. Each sorts of uncertainty are able to generalizing calibration underneath distribution shift. We additionally present proof that GPT-3’s skill to generalize calibration is determined by pre-trained latent representations that correlate with epistemic uncertainty over its solutions.
