Benchmarks Don't Talk Back: Learning Through Self-Disagreement A November 2026 arXiv paper (2611.17563v1) introduces "dialectical competence" — a model's ability to hold a correct position or yield to a better one — and a training method called Self-Disagreement that teaches it without a second model, reward model, preference data or human annotation. In the method, a model states a position, generates the strongest case against it, rules on whether that case stands, and is trained via reinforcement learning from verifiable rewards, with the same reference check scoring both the answer and the ruling. The authors report the capability is weak in the base model, emerges with the reward, does not cost first-answer accuracy, and holds against challengers the model never trained against, including stronger ones. arXiv:2611.17563v1 Announce Type: new Abstract: A benchmark usually measures a language model alone with a question, and deployment rarely looks like that: answers get contested, and models give up correct ones under counter-reasoning that is itself wrong. We show that what fails there is not a shortfall of accuracy but a capability in its own right -- knowing when to hold a position and when to yield to a better one -- which we call dialectical competence. It stands on first-answer accuracy without following from it: a model can be made materially better at answering and no better at this, and training for accuracy alone can drift against it. Nor can it be bought at inference, by self-reflection or by sampling more. Taking a cue from the dialectical tradition in philosophy, in which productive opposition arises within a position rather than being imposed on it from outside, we hypothesize that this capability can be trained from opposition the model supplies itself. It states a position, generates the strongest case against that position -- its antithesis -- rules on whether the case stands, and is trained on that ruling, with no second model at training or at deployment. We call this learning through Self-Disagreement: reinforcement learning from verifiable rewards throughout, where the same check that scores an answer against its reference scores the ruling too, so no reward model, no preference data and no human annotation enter anywhere. The capability is trained rather than elicited: weak in the base model, it arrives with the reward, is not paid for out of first-answer accuracy, and holds against challengers the model never trained against, including ones stronger than itself. The opposition a model needs in order to improve, it turns out, need not come from outside it.