Skip to content
Last Dictionary
Plane
Training & Post-trainingLanguage Models & TransformersSymbolic & Classical AIEvaluation & Benchmarks

verifier

nounAlso: grader, checker, verification model

In plain wordsA program or AI model that checks whether a proposed answer is correct — by running tests, comparing with a known result or judging the reasoning — so that good answers can be picked or rewarded.

  1. 1

    Language models

    A component that assesses a generator's candidate outputs for correctness, used to select among samples at inference time (best-of-n samplingbest-of-n sampling, n.: Asking an AI the same question several times and keeping whichever answer a scoring model or checker rates highest — a simple way to trade extra computing for better answers., search) or to supply rewards in training. Verifiers range from exact checkers — answer matching against a reference, test execution, compilers, proof assistants — to learned models: outcomeoutcome reward model, n.: A model that judges only whether an AI's final answer is right, ignoring how it got there — like a teacher who checks just the answer box. and processprocess reward model, n.: A model that grades each step of an AI's worked solution, not just the final answer, so mistakes can be caught where they happen — like a teacher marking every line of a proof. reward models that output a probability of correctness, and generative verifiers that reason in natural language before giving a verdict. Their value rests on the observation that checking is often easier than solving, so a verifier can lift performance beyond the generator's single-attempt accuracy (a difference sometimes called the generation–verification gap). Exact verifiers enable reinforcement learning with verifiable rewardsreinforcement learning with verifiable rewards, n.: Training an AI by letting it attempt problems whose answers can be checked automatically — a math result, a passing program — and rewarding it only when the check succeeds.; learned ones can be fooled, so optimizing hard against them invites reward overoptimizationreward overoptimization, n.: What happens when an AI is trained too hard to please an imperfect scoring model: its scores keep climbing while the real quality of its answers stops improving and then gets worse..

    Our verifier runs the hidden test suite, so the policy only gets credit for code that actually passes.

  2. 2

    Formal methods and complexity

    A procedure that checks a purported proof or certificate: a proof assistant's kernel checking a formal proof, or the efficient checker in an interactive proof system, which accepts a more powerful prover's claims only if they survive its tests. Formal verifiers such as the Lean kernel let models be trained and evaluated on mathematics with guaranteed-correct feedback, and prover–verifier games train models to produce solutions that weaker checkers can confirm.

Origin

The complexity-theory sense comes from interactive proofs (Shafi Goldwasser, Silvio Micali and Charles Rackoff; László Babai; both 1985). In language modeling, Karl Cobbe and colleagues at OpenAI (2021) trained verifiers to rank sampled solutions to GSM8K problems, establishing the generate-then-verify pattern behind later reward models and reasoning systems.

Usage note

Usage overlaps with reward modelreward model, n.: A separate AI model trained to score responses the way people would, so it can stand in for human judges while another model is trained on millions of examples. and LLM-as-a-judgeLLM-as-a-judge, n.: Using one AI language model to grade or compare the answers of another, as a cheaper and faster stand-in for human reviewers.: 'verifier' stresses correctness, 'reward model' preference, and 'judge' evaluation. A 'verifiable' task is one with an automatic checker, not necessarily a formal proof.

Related

See also
  • outcome reward modeloutcome reward model, n.: A model that judges only whether an AI's final answer is right, ignoring how it got there — like a teacher who checks just the answer box.
  • process reward modelprocess reward model, n.: A model that grades each step of an AI's worked solution, not just the final answer, so mistakes can be caught where they happen — like a teacher marking every line of a proof.
  • reward modelreward model, n.: A separate AI model trained to score responses the way people would, so it can stand in for human judges while another model is trained on millions of examples.
  • best-of-n samplingbest-of-n sampling, n.: Asking an AI the same question several times and keeping whichever answer a scoring model or checker rates highest — a simple way to trade extra computing for better answers.
  • reinforcement learning with verifiable rewardsreinforcement learning with verifiable rewards, n.: Training an AI by letting it attempt problems whose answers can be checked automatically — a math result, a passing program — and rewarding it only when the check succeeds.
  • LLM-as-a-judgeLLM-as-a-judge, n.: Using one AI language model to grade or compare the answers of another, as a cheaper and faster stand-in for human reviewers.
  • interactive theorem provinginteractive theorem proving, n.: Writing a proof in a precise computer language so that software checks every step: the person supplies the ideas, the computer fills in small steps and guarantees that nothing is wrong.
Contrast with
  • reward modelreward model, n.: A separate AI model trained to score responses the way people would, so it can stand in for human judges while another model is trained on millions of examples.
  • LLM-as-a-judgeLLM-as-a-judge, n.: Using one AI language model to grade or compare the answers of another, as a cheaper and faster stand-in for human reviewers.

Sources

  1. Cobbe, K. et al. (2021). Training Verifiers to Solve Math Word Problems. arXiv:2110.14168. arxiv.org/abs/2110.14168
  2. Goldwasser, S., Micali, S. & Rackoff, C. (1985). The Knowledge Complexity of Interactive Proof-Systems. Proc. 17th ACM Symposium on Theory of Computing.

Mentioned in

best-of-n samplingbest-of-n sampling, n.: Asking an AI the same question several times and keeping whichever answer a scoring model or checker rates highest — a simple way to trade extra computing for better answers., formal verificationformal verification, n.: Using mathematics, not just testing, to prove that a computer system always behaves as specified — for example, that a chip never adds wrongly or that a program can never crash in a particular way., GSM8KGSM8K, prop. n.: A set of about 8,500 grade-school math word problems, each needing a few steps of simple arithmetic, used to test whether language models can reason through a problem step by step., LLM-as-a-judgeLLM-as-a-judge, n.: Using one AI language model to grade or compare the answers of another, as a cheaper and faster stand-in for human reviewers., reinforcement fine-tuningreinforcement fine-tuning, n.: Improving a trained AI model by trial and error instead of examples to copy: it tries answers, a grader scores them, and the model shifts toward whatever earned high scores., reinforcement learning with verifiable rewardsreinforcement learning with verifiable rewards, n.: Training an AI by letting it attempt problems whose answers can be checked automatically — a math result, a passing program — and rewarding it only when the check succeeds., rejection samplingrejection sampling, n.: Generating many candidate answers and keeping only those that pass a check — a scoring model, a test or a correct final answer — often so the AI can then be trained on its own best work., reward modelreward model, n.: A separate AI model trained to score responses the way people would, so it can stand in for human judges while another model is trained on millions of examples., RL environmentRL environment, n.: A practice world built for training an AI by trial and error: it hands the AI tasks and tools, lets it act, and scores the results so the AI can learn what works., self-correctionself-correction, n.: An AI model noticing and fixing its own mistakes — by rereading its answer, running a test or checking a source — before it gives a final result., self-improvementself-improvement, n.: When an AI gets better by learning from material it produces itself — its own answers, critiques or practice problems — rather than only from new human-made examples., test-time computetest-time compute, n.: The computing effort an AI model spends while answering a question, as opposed to while being trained. Letting a model think longer or try many answers can make it markedly smarter., evaluator–optimizerevaluator–optimizer, n.: A pairing in which one AI produces a draft and another AI grades it and suggests fixes, repeating the cycle until the work is good enough., expert iterationexpert iteration, n.: A learning loop in which a slow, careful search finds better moves or solutions than the AI's quick instincts, and the AI is then trained to imitate them, so each round's search starts smarter., generative reward modelgenerative reward model, n.: A scoring model that judges an AI's answer by writing out its reasoning in words before giving a verdict, instead of simply outputting a number., guaranteed safe AIguaranteed safe AI, n.: A proposed approach to AI safety in which systems come with mathematical proofs or checkable guarantees that they will not cause certain harms, instead of relying on testing alone., outcome reward modeloutcome reward model, n.: A model that judges only whether an AI's final answer is right, ignoring how it got there — like a teacher who checks just the answer box., process reward modelprocess reward model, n.: A model that grades each step of an AI's worked solution, not just the final answer, so mistakes can be caught where they happen — like a teacher marking every line of a proof., prover–verifier gameprover–verifier game, n.: A training setup in which one AI must justify its answers so clearly that a weaker AI can check them, which also makes the strong AI's work easier for people to check., ReflexionReflexion, prop. n.: A technique in which an AI agent that fails at a task writes itself a short note on what went wrong, then reads that note before trying again — learning from mistakes without retraining., REINFORCE leave-one-outREINFORCE leave-one-out, n.: A simple reinforcement-learning method that has a model answer each question several times and judges each answer against the average score of the other answers., self-taught reasonerself-taught reasoner, n.: A training method in which a language model writes step-by-step explanations for problems, keeps the ones that led to correct answers, and learns from them, so it gradually teaches itself to reason., stop conditionstop condition, n.: The rule that tells an AI agent when to stop working — because the task is done, a time or cost limit has been reached, or something has gone wrong.