Epistemic Constitutionalism Or: how to avoid coherence bias

arXiv:2601.14295v3 Announce Type: replace Abstract: Large language models increasingly function as artificial reasoners: they evaluate arguments, assign credibility, and express confidence. Yet their belief-forming behavior is governed by implicit, uninspected epistemic policies. This paper argues...