Instruction Following (IF) is a core capability of LLMs, requiring strict adherence to diverse constraints, ranging from verifiable ones (e.g., output length) to unverifiable ones (e.g., tone). Reinforcement learning with verifiable rewards has emerged as a paradigm for IF tasks, leveraging LLM-as-a-judge to assess unverifiable constraints. However, we empirically find that this approach remains a significant bottleneck, suffering from severe reward hacking and higher computational overhead. In this work, we first analyze the generalization capabilities of unverifiable constraints and discover that specific constraints exhibit distinct, high-generalization patterns. Motivated by this, we propose TinyJudge, a framework that employs an ensemble of specialized tiny language models ($\sim0.6B$) to provide rewards for soft constraints. By distilling expertise from frontier models into these tiny models, it achieves high-precision, lightweight evaluation. Extensive evaluations across five benchmarks demonstrate that TinyJudge outperforms the baselines by $\sim10\%$ in average performance and $12\%$ in reward precision. Crucially, it also achieves a $3\times$ speedup in total training tim
Reinforcement Learning with Verifiable Rewards (RLVR) has become a cornerstone for unlocking complex reasoning in Large Language Models (LLMs). Yet, scaling up RL is bottlenecked by limited existing verifiable data, where improvements increasingly saturate over prolonged training. To overcome this, we propose Golden Goose, a simple trick to synthesize unlimited RLVR tasks from unverifiable internet text by constructing a multiple-choice question-answering version of the fill-in-the-middle task. Given a source text, we prompt an LLM to identify and mask key reasoning steps, then generate a set of diverse, plausible distractors. This enables us to leverage reasoning-rich unverifiable corpora typically excluded from prior RLVR data construction (e.g., science textbooks) to synthesize GooseReason-0.7M, a large-scale RLVR dataset with over 0.7 million tasks spanning mathematics, programming, and general scientific domains. Empirically, GooseReason effectively revives models saturated on existing RLVR data, yielding robust, sustained gains under continuous RL and achieving new state-of-the-art results for 1.5B and 4B-Instruct models across 15 diverse benchmarks. Finally, we deploy Golden
The prevailing paradigm for training large reasoning models--combining Supervised Fine-Tuning (SFT) with Reinforcement Learning with Verifiable Rewards (RLVR)--is fundamentally constrained by its reliance on high-quality, human-annotated reasoning data and external verifiers. This dependency incurs significant data-collection costs, risks embedding human cognitive biases, and confines the reinforcement learning stage to objectively assessable domains like mathematics and coding, leaving a wide range of unverifiable tasks beyond its scope. To overcome these limitations, we introduce NRT (Native Reasoning Training), a novel framework that cultivates complex reasoning by having the model generate its own reasoning traces using only standard question-answer pairs, thereby obviating the need for expert-written demonstrations. NRT reframes the training problem by treating the reasoning process as a latent variable. It employs a unified training objective that models reasoning as an optimization problem, intrinsically rewarding paths that increase the model's likelihood of producing the ground-truth answer. This unified perspective allows us to analyze intrinsic failure modes of prior met
We propose to scale RL to unverifiable data with a novel algorithm JEPO (Jensen's Evidence lower bound Policy Optimization). While most prior efforts on scaling RL for LLMs focus on verifiable data where ground truth answers are typically short-form and can be matched easily; we investigate the case where such assumptions are less valid (e.g., when answers are long-form such as mathematical proofs). To scale RL training to unverifiable data with contemporary training constraints, we propose JEPO. JEPO applies Jensen's evidence lower bound, a pragmatic simplification of the evidence lower bound which views chain-of-thought as a latent variable in the generative process. We show that on verifiable data (math), JEPO is as effective as RL with verifiable rewards; on semi-verifiable data (numina), JEPO improves on soft-match based evaluations compared to RL with verifiable rewards which can only leverage a subset of the data source; finally, on unverifiable data (numina-proof), JEPO outperforms SFT and a few ablation baselines on likelihood evaluations.
The rising cost of acquiring supervised data has driven significant interest in self-improvement for large language models (LLMs). Straightforward unsupervised signals like majority voting have proven effective in generating pseudo-labels for verifiable tasks, while their applicability to unverifiable tasks (e.g., translation) is limited by the open-ended character of responses. As a result, self-evaluation mechanisms (e.g., self-judging and entropy minimization) are predominantly used to derive pseudo-labels. However, self-evaluation relying on LLMs typically incurs high computational overhead and introduces overconfidence issues due to intrinsic biases. To address these challenges, we propose a novel self-evaluation-free approach for unverifiable tasks, designed for lightweight yet effective self-improvement. Inspired by majority voting commonly employed in verifiable tasks, we propose semantic voting as a novel mechanism that relaxes the principle of hard matching (i.e., exact matching) toward soft matching (i.e., semantic similarity). Soft matching is achieved by leveraging a lightweight sentence embedding model to quantify semantic similarity, thereby mitigating excessive comp
Large Language Model (LLM) agents can increasingly automate complex reasoning through Test-Time Scaling (TTS), iterative refinement guided by reward signals. However, many real-world tasks involve multi-stage pipeline whose final outcomes lack verifiable rewards or sufficient data to train robust reward models, making judge-based refinement prone to accumulate error over stages. We propose Selective TTS, a process-based refinement framework that scales inference across different stages in multi-agent pipeline, instead of repeated refinement over time by prior work. By distributing compute across stages and pruning low-quality branches early using process-specific judges, Selective TTS mitigates the judge drift and stabilizes refinement. Grounded in the data science pipeline, we build an end-to-end multi-agent pipeline for generating visually insightful charts and report of given dataset, and design a reliable LLM-based judge model, aligned with human experts (Kendall's τ=0.55). Our proposed selective TTS then improves insight quality under a fixed compute budget, increasing mean scores from 61.64 to 65.86 while reducing variance. We hope our findings serve as the first step toward
Reinforcement learning (RL) training of large language models (LLMs) on unverifiable tasks is challenging even when a reasonable-quality reference answer is available. We propose a constrained RL training framework that (i) optimizes a token-level dense Reasoning Reflection Reward (R3) aligned with reasoning quality, and (ii) enforces rubric-gating as feasibility constraints at the rollout group level. R3 measures the model's token-level certainty of a reference answer under its chain-of-thought (CoT) prefix, and selectively emphasizes tokens with high cross-rollout variance, which we call reasoning-reflective tokens, that would otherwise be diluted by the bulk of low-variance tokens. The same variance signal also drives a filter that discards queries with insufficient signal for comparative learning. Rubric-gating complements R3 by operationalizing principled task criteria as hard accept/reject checks on final answers. Empirically, across four datasets spanning scientific writing, medicine, legal contracts, and finance, our framework outperforms strong baselines, achieves faster, more sample-efficient learning, and respects feasibility constraints.
Prediction markets elicit and aggregate beliefs by paying agents based on how close their predictions are to a verifiable future outcome. However, outcomes of many important questions are difficult to verify or unverifiable, in that the ground truth may be hard or impossible to access. We present a novel incentive-compatible prediction market mechanism to elicit and efficiently aggregate information from a pool of agents without observing the outcome, by paying agents the negative cross-entropy between their prediction and that of a carefully chosen reference agent. Our key insight is that a reference agent with access to more information can serve as a reasonable proxy for the ground truth. We use this insight to propose self-resolving prediction markets that terminate with some probability after every report and pay all but a few agents based on the final prediction. The final agent is chosen as the reference agent since they observe the full history of market forecasts, and thus have more information by design. We show that it is a perfect Bayesian equilibrium (PBE) for all agents to report truthfully in our mechanism and to believe that all other agents report truthfully. Altho
In future, information may become one of the most important assets in economy. However, unlike common goods (e.g. clothing), information is troublesome in trading since the information commodities are \emph{vulnerable}, as they lose their values immediately after revelation, and possibly unverifiable, as they can be subjective. By authorizing a trusted center (e.g. Amazon) to help manage the information trade, traders are ``forced'' to give the trusted center the ability to become an information monopolist. To this end, we need a trust-free (i.e. without a trusted center and with only strategic traders) unverifiable information trade protocol such that it 1) motivates the sellers to provide high quality information, and the buyer to pay for the information with a fair price (truthful); 2) except the owner, the information is known only to its buyer if the trade is executed (secure). In an unverifiable information trade scenario (e.g. a medical company wants to buy experts' opinions on multiple difficult medical images with unknown pathological truth from several hospitals), we design a trust-free, truthful, and secure protocol, Smart Info-Dealer (SMind), for information trading, by
Agents powered by large language models (LLMs) are increasingly deployed in settings where communication shapes high-stakes decisions, making a principled understanding of strategic communication essential. Prior work largely studies either unverifiable cheap-talk or fully verifiable disclosure, failing to capture realistic domains in which information has probabilistic credibility. We introduce MixTalk, a strategic communication game for LLM-to-LLM interaction that models information credibility. In MixTalk, a sender agent strategically combines verifiable and unverifiable claims to communicate private information, while a receiver agent allocates a limited budget to costly verification and infers the underlying state from prior beliefs, claims, and verification outcomes. We evaluate state-of-the-art LLM agents in large-scale tournaments across three realistic deployment settings, revealing their strengths and limitations in reasoning about information credibility and the explicit behavior that shapes these interactions. Finally, we propose Tournament Oracle Policy Distillation (TOPD), an offline method that distills tournament oracle policy from interaction logs and deploys it in
The tendency for hallucination in current large language models (LLMs) negatively impacts dialogue systems. Such hallucinations produce factually incorrect responses that may mislead users and undermine system trust. Existing refinement methods for dialogue systems typically operate at the response level, overlooking the fact that a single response may contain multiple verifiable or unverifiable facts. To address this gap, we propose Fine-Refine, a fine-grained refinement framework that decomposes responses into atomic units, verifies each unit using external knowledge, assesses fluency via perplexity, and iteratively corrects granular errors. We evaluate factuality across the HybriDialogue and OpendialKG datasets in terms of factual accuracy (fact score) and coverage (Not Enough Information Proportion), and experiments show that Fine-Refine substantially improves factuality, achieving up to a 7.63-point gain in dialogue fact score, with a small trade-off in dialogue quality.
Reinforcement learning with verifiable rewards (RLVR) enhances the reasoning of large language models (LLMs), but standard RLVR often depends on human-annotated answers or carefully curated reward specifications. In machine-checkable domains, label-free alternatives such as majority voting or LLM-as-a-judge remove annotation cost but can introduce false positives that destabilize training. We introduce JURY-RL, a label-free RLVR framework that decouples answer proposal from reward disposal: votes from model rollouts propose a candidate answer, and a formal verifier determines whether that candidate can receive positive reward. Concretely, only rollouts matching the plurality-voted answer are rewarded when that answer is successfully verified in Lean. When verification is inconclusive, we invoke ResZero (Residual-Zero), a fallback reward that discards the unverified plurality proposal and redistributes a zero-mean, variance-preserving signal over the residual answers. This design maintains a stable optimization gradient without reinforcing unverifiable consensus. Across three backbone models trained on mathematical data, JURY-RL consistently outperforms other label-free baselines on
The psychological profile that structurally documents the case of a depression patient is essential for psychotherapy. Large language models can be applied to summarize the profiles from counseling speech, however, it may suffer from long-context forgetting and produce unverifiable hallucinations, due to overlong length of speech, multi-party interactions and unstructured chatting. Hereby, we propose a StreamProfile, a streaming framework that processes counseling speech incrementally, extracts evidences grounded from ASR transcriptions by storing it in a Hierarchical Evidence Memory, and then performs a Chain-of-Thought pipeline according to PM+ psychological intervention for clinical reasoning. The final profile is synthesized strictly from those evidences, making every claim traceable. Experiments on real-world teenager counseling speech have shown that the proposed StreamProfile system can accurately generate the profiles and prevent hallucination.
The growing publication rate of research papers has created an urgent need for better ways to fact-check information, assess writing quality, and identify unverifiable claims. We present Plato's Cave as an open-source, human-centered research verification system that (i) creates a directed acyclic graph (DAG) from a document, (ii) leverages web agents to assign credibility scores to nodes and edges from the DAG, and (iii) gives a final score by interpreting and evaluating the paper's argumentative structure. We report the system implementation and results on a collected dataset of 104 research papers.
Causal Discovery (CD) is a powerful framework for scientific inquiry. Yet, its practical adoption is hindered by a reliance on strong, often unverifiable assumptions and a lack of robust performance assessment. To address these limitations and advance empirical CD evaluation, we present TCD-Arena, a modularized, highly customizable, and extendable testing kit to assess the robustness of time series CD algorithms against stepwise more severe assumption violations. For demonstration, we conduct an extensive empirical study comprising around 30 million individual CD attempts and reveal nuanced robustness profiles for 33 distinct assumption violations. Further, we investigate CD ensembles and find that they have the potential to improve general robustness, which has implications for real-world applications. With this, we strive to ultimately facilitate the development of CD methods that are reliable for a diverse range of synthetic and potentially real-world data conditions.
Hallucination--defined here as generating statements unsupported or contradicted by available evidence or conversational context--remains a major obstacle to deploying conversational AI systems in settings that demand factual reliability. Existing metrics either evaluate isolated responses or treat unverifiable content as errors, limiting their use for multi-turn dialogue. We introduce VISTA (Verification In Sequential Turn-based Assessment), a framework for evaluating conversational factuality through claim-level verification and sequential consistency tracking. VISTA decomposes each assistant turn into atomic factual claims, verifies them against trusted sources and dialogue history, and categorizes unverifiable statements (subjective, contradicted, lacking evidence, or abstaining). Across eight large language models and four dialogue factuality benchmarks (AIS, BEGIN, FAITHDIAL, and FADE), VISTA substantially improves hallucination detection over FACTSCORE and LLM-as-Judge baselines. Human evaluation confirms that VISTA's decomposition improves annotator agreement and reveals inconsistencies in existing benchmarks. By modeling factuality as a dynamic property of conversation, VI
We establish the mathematical limits of AGI safety in two forms: verifying a fixed system, and verifying that a certified safety property persists once the system self-modifies. In the static case, no algorithm can certify a highly expressive AGI's safe behaviour infallibly, completely and tractably, whether over unbounded input domains (blocked by Rice's and Godel's theorems) or over all finite hardware configurations (blocked by Trakhtenbrot's theorem, which splits into a PSPACE-hardness barrier and a co-RE-completeness barrier), forcing a Soundness-Completeness-Tractability Trilemma as a structural, not statistical, necessity. In the dynamic case, we formalise self-modification as a computable transition operator and prove that no algorithm can determine, from a system's current certified safety, whether safety survives its next self-modification step: a result that reduces to Rice's Theorem one level up, making the static and dynamic barriers two faces of one obstruction. This forces an exclusive dichotomy: persistent certification is attainable only for systems that have stopped evolving semantically, i.e. only for narrow, not general, systems. Nor can the obstruction be deleg
The April 2026 Claude Mythos sandbox escape exposed a critical weakness in frontier AI containment: the infrastructure surrounding advanced models remains susceptible to formally characterizable arithmetic vulnerabilities. Anthropic has not publicly characterized the escape vector; some secondary accounts hypothesize a CWE-190 arithmetic vulnerability in sandbox networking code. We treat this as unverified and analyze the vulnerability class rather than the specific escape. This paper presents COBALT, a Z3 SMT-based formal verification engine for identifying CWE-190/191/195 arithmetic vulnerability patterns in C/C++ infrastructure prior to deployment. We distinguish two classes of contribution. Validated: COBALT detects arithmetic vulnerability patterns in production codebases, producing SAT verdicts with concrete witnesses and UNSAT guarantees under explicit safety bounds. We demonstrate this on four production case studies: NASA cFE, wolfSSL, Eclipse Mosquitto, and NASA F Prime, with reproducible encodings, verified solver output, and acknowledged security outcomes. Proposed: a four-layer containment framework consisting of COBALT, VERDICT, DIRECTIVE-4, and SENTINEL, mapping pre-
Standard procurement models assume that the buyer knows the quality of the good at the time of procurement; however, in many settings, the quality is learned only long after the transaction. We study procurement problems in which the buyer's valuation of the supplied good depends directly on its quality, which is unverifiable and unobservable to the buyer. For a broad class of procurement problems, we identify procurement mechanisms maximizing any weighted average of the buyer's expected payoff and social surplus. The optimal mechanism can be implemented via an auction that restricts sellers to submitting bids within specific intervals.
Reasoning in language models is difficult to evaluate: natural-language traces are unverifiable, symbolic datasets are too small, and most benchmarks conflate heuristics with inference. We present FOL-Traces, the first large-scale dataset of programmatically verified reasoning traces, enabling rigorous evaluation of structured logical inference. We also propose two challenging and comprehensive diagnostic tasks-masked operation prediction and step completion-that directly probe syntactic awareness and process fidelity. FOL-Traces serves as a scalable testbed for rigorously studying how models perform structured logical inference. Systematic experiments with 5 reasoning LLMs show that the dataset remains challenging: models only reach around 45.7% accuracy on masked operation prediction and around 27% on two-step completion.