arXiv subjects

Tianshi Zheng

Publications and source records attributed to Tianshi Zheng.

At least 19 recordsLinked to original sources

arXiv2026

Toward Knowledge-Centric AI: Discovering, Internalizing, and Using Explicit Knowledge Across Agents and Generations

Knowledge is how understanding accumulates across people and generations: durable records that outlive their producers and that others can check, combine, teach, and correct. AI agents are now built and run by many independent users and organizations, and each agent is upgraded, replaced, or retired within months. Yet what an agent learns stays trapped in model parameters, a transient context window, or a system-specific memory store that is appended to and retrieved from rather than processed. We argue that the next generation of AI should be knowledge-centric, built around an explicit, continuously evolving knowledge layer independent of any single agent. In this loop, agents discover knowledge from their environments and trajectories; the layer internalizes it by turning fragmented, heterogeneous records and experience into composable, revisable understanding; and agents across systems and generations use the result and return new outcomes for revision. While discovery and use have advanced rapidly with large language model agents, internalization remains the bottleneck. We argue that internalization should become a persistent object of training and maintenance-centered on condition-preserving conceptualization of events and deductive, inductive, and abductive reasoning across noisy, heterogeneous representations (text, tables, graphs, rules, and programs). To power this layer, we call for knowledge foundation models that provide shared neural capacity for knowledge computation, and outline a research agenda showing how an explicit knowledge layer delivers reuse, cost efficiency, auditability, and continuity across agent generations.

cs.AI↗

arXiv2026

SCI-AGENT-EXAM: Benchmarking Agents on Scientific Research Tasks

Large language model (LLM) agents are increasingly capable of supporting scientific workflows, but their capabilities across diverse research activities remain difficult to assess systematically. Existing benchmarks often focus on selected scientific capabilities or task types, limiting a comprehensive characterization of scientific agents' strengths and weaknesses in realistic research workflows. We introduce SCI-AGENT-EXAM, a workflow-oriented benchmark comprising 120 expert-curated tasks across 11 scientific fields and 7 research task types defined by the Scientific Workflow Task Taxonomy (SWTT). Each task requires both executable artifacts (e.g., code and scripts) and textual artifacts (e.g., reports and explanations). To evaluate the heterogeneous task outputs, we develop a hybrid evaluation protocol combining deterministic programmatic verification for executable artifacts with a calibrated, expert-validated LLM-based judge for textual artifacts. Beyond outcome assessment, we analyze progress and solution stability along execution trajectories to characterize how agents reach their final results. Experiments with 10 frontier LLM agents reveal substantial performance gaps: GPT-6 Astra achieves the highest overall score of 80.9, yet only 26.0% of its runs exceed the 0.95 success threshold. Further analyses uncover distinct capability profiles across scientific task types and substantial variation in execution efficiency and stability. Our code and data are available at https://github.com/SKYLENAGE-AI/Sci-Agent-Exam.

cs.CL↗

arXiv2026

AdaExploreBench: Benchmarking LLM Agents for Proactive Exploration and Test-time Adaptation in Evolving Environments

Large Language Model (LLM) agents are increasingly deployed in unfamiliar and evolving environments, requiring them to proactively explore, understand, and adapt to environmental dynamics. However, existing benchmarks rarely test whether agents can proactively discover how an environment works and adapt as it evolves. To address this gap, we introduce AdaExploreBench, a benchmark for proactive exploration and test-time adaptation in evolving environments. By instantiating environmental dynamics as hidden rules, AdaExploreBench requires agents to actively acquire evidence and infer the rules. Under implicit rule drift, the latent rule changes without announcement, requiring agents to detect from degraded performance that prior knowledge is stale and re-explore. Experiments on eight leading LLMs show that proactive and adaptive exploration remain challenging: GPT-6 Sol achieves only 61.9% calibrated accuracy, while providing sufficient evidence or removing rule drift improves average performance significantly by close to 10%. These results reveal a critical bottleneck in current LLM agents: they can often reason effectively from informative evidence once it is available, but struggle to actively acquire the right evidence and recognize when previously acquired evidence is no longer valid.

cs.CL↗

arXiv2026

CoworkEnv: Synthesizing Training Environments for Long-Horizon Cowork Agents

AI coworkers must coordinate tools, communicate across turns, and act on changing workspace state while respecting workflow constraints. Constructing training experience for these capabilities requires coherent requests, service records, tool interfaces, and grading checks. We present CoworkEnv, a capability-guided framework for synthesizing executable training environments for AI coworkers. Workflow requirements guide task construction, and programmatic validation with feedback repair checks generated specifications and grading behavior. Teacher executions yield curated demonstrations for supervised fine-tuning (SFT), while student interactions provide rollouts and rewards for reinforcement learning (RL). Using these resources, we train compact PocketWorker models. For Qwen3-VL-8B-Instruct, SFT raises the mean score from 46.6 to 64.1 on Claw-Eval, including a 31.9-point gain on Multi-turn tasks. An SFT-only extension to ClawMark demonstrates applicability across coworker benchmarks. We further compare RL training dynamics across configurations that vary reward processing and credit assignment. Improving training rewards provide preliminary evidence that the constructed environments support policy optimization.

cs.AI↗

arXiv2026

When Should an In-Context Learner Expand Its Hypothesis Space?

Learning systems adapt quickly inside a familiar family of models. The harder step comes earlier: deciding, from observations that could be noise, an exception, a change within the family or structure outside it, whether opening a richer family is worth its cost. We treat this as a costly sequential decision: prediction failure must be turned into structural evidence, evidence into a value of expansion, and value into action. The Structural Revision Environment produces matched failures from each source, varies the price of expansion and the remaining horizon independently of the evidence, and admits exact Bayesian calculations and an exact normative solution of the one-shot decision. Its solution shows that revision is a value boundary and not an evidence threshold: one history has different optimal actions under different prices, horizons and announced queries, the boundary between local repair and expansion is set by the inputs a rule predicts and a repair cannot cover, and belief in the richer family crosses long before the decision does. Transformers trained in the environment reproduce this boundary from utility alone. Language models of three post-training lineages carry a failure-sensitive signal in their predictions that is not reflected in their revision decisions, and given the gain of expanding they read it without weighing it against price and horizon. Three models allowed to reason weigh the stated gain in the reference's proportions and still do not turn the history into an estimate of what expansion would buy. Controlled post-training of the meta-trained learners moves the prior and the sharpness of predictions, and neither moves the criterion.

cs.LG↗

arXiv2026

ULTRADISCOVERY: Abductive Exploration in an Interconnected, Epistemically Open Universe

Scientific discovery often begins when scattered clues call for a new way of describing the world. Such abductive exploration can require constructing the representation in which an explanation is stated, when the world is epistemically open, and composing evidence scattered across contexts, when it is structurally interconnected. Existing benchmarks rarely separate these two demands or control them independently. We introduce ULTRADISCOVERY, an interactive world of five domains in which an agent revises an initially successful theory and predicts the outcome of an unseen cross-domain intervention. A $2 \times 2$ design leaves the representation open or discloses it, and leaves the evidence distributed or aligns it, with the latent dynamics fixed. With the representation open, agents across eleven models often retract the axiom they were taught, and none introduces the unobserved entity or rewrites the variables that a replacement requires. Disclosure triples intervention requests and adds about one of the eighteen findings the world affords, and alignment adds less. Two vendor-harness systems carry discovery into more domains, and one of them rewrites the variables in Open episodes. No system makes the exact prediction within 200 paid actions. At larger budgets one exact prediction appears with both aids, while every Open episode remains inexact. The results locate the difficulty in the step from accumulating evidence to composing it into a representation that transfers.

cs.LG↗

arXiv2026

Same Outcome, Different Readout: What Does a Steerable Valence Direction in LLMs Represent?

Claims about what an internal direction in an LLM represents need evidential constraints beyond an observer's prior beliefs about the system. Decodability and successful activation steering do not, by themselves, establish which construct the direction tracks. This gap is especially consequential for welfare-relevant interpretations, where a proposed functional state must be distinguished from correlated features of the extraction contrast. We treat the question as one of construct validity and study a good-bad outcome direction in a maze task, using controlled interventions that separate the realised outcome from the informational history through which it became known. Across multiple LLM checkpoints, directions fitted on one explicit outcome encoding transfer well to another, indicating that the readout is not tied to surface form. When the same realised outcome is reached through announced and unannounced histories, however, transfer degrades substantially: even after both histories receive the same explicit outcome, the post-event readout remains strongly conditioned on the earlier announcement. In the base model, steering along the direction changes actions, yet removing it leaves the natural cue effect almost intact. In a matched maze-RL run, the post-RL direction becomes substantially more predictive of reference-MDP remaining return and the policy becomes more dependent on it at the tested sites, while the history dependence persists. These dissociations support a functional, value-related interpretation of the direction, but not its identification with a history-invariant scalar valence state.

cs.LG↗

arXiv2026

SkillHorizon: Constructing Long-Horizon Agent Tasks through Skill Composition

Long-horizon agents must complete workflows in which later activities depend on earlier outputs and changes to shared state. A key challenge for evaluation is to expand task requirements and their dependencies while keeping tasks executable and outcomes verifiable. We introduce SkillHorizon, a benchmark that constructs long-horizon tasks through skill composition in seven business environments. Its generators use domain-specific rules to turn reusable skill procedures into concrete business activities. These activities are connected through input-output dependencies, shared resources, and state updates, allowing workload and coordination demands to grow together. Successful reference executions validate task feasibility and measure interaction workload, while programmatic evaluators assess local completion and whole-task success. We evaluate five language models on 1,050 tasks with reference interaction workloads ranging from 64K to 1M tokens. Across these scales, mean local-goal completion falls from 90.4% to 46.8%, as completed work fails to keep pace with growing requirements. At larger workloads, agents more often submit unsuccessful final answers with substantial work unfinished and tool-call allowance remaining. Trace analysis also identifies cases where scripts execute successfully but their outputs fail downstream business checks. SkillHorizon provides a configurable testbed for studying sustained progress, dependency coordination, and complete task delivery in growing workflows. Our code is available at https://github.com/KnowComp/SkillHorizon.

cs.AI↗

arXiv2026

Erased, Rerouted, or Rescaled? Post-Training and the Causal Quotient of a Language Model's Belief State

What happens to information a pretrained model already encodes when post-training no longer rewards using it? The common language of representation compression conflates three fates: information may be erased, rerouted away from the decision while still represented, or rescaled to occupy less variance while still represented and used. We make these fates identifiable in models whose pretraining recovers Bayesian belief states. A reward that reads only a coarse function of the hidden state defines an exact reward-null kernel. The kernel lets us separately measure whether the information remains recoverable, whether decisions causally depend on it, and how much activation variance it occupies. Theory says what is protected: KL-anchored reinforcement learning preserves the reference policy's log-odds among equally rewarded outputs, supervised and unanchored objectives carry no such constraint, and spectral compression implies neither erasure nor loss of use. In controlled worlds, post-training mostly reroutes or rescales reward-null information and leaves it decodable. Without an anchor decisions can stop using it although the representation survives, and with one they keep using it. Erasure appears only under prolonged weight decay, for distinctions that neither reward nor next-token prediction can see. Open language models show the same dissociation: in-context belief geometry stays decodable under late-layer spectral compression, and within-class behavior depends on the anchor. Post-training thus selects a causal quotient of the pretrained belief state: the reward defines decision-equivalence, the anchor and the state update protect part of what it ignores, and optimization decides whether the rest is erased, rerouted, or rescaled.

cs.LG↗

arXiv2026

One Hypothesis Is Not Enough: Abductive Reasoning with Agentic Hypothesis Refinement over Knowledge Graphs

Abductive reasoning over knowledge graphs (KGs) seeks a first-order logic hypothesis whose answer set explains a given set of observed entities. Since many hypotheses can explain the same observations, controllable hypothesis generators condition generation on entities, relations, or logical patterns, but they treat generation as a single step. A generated hypothesis may be well-formed and satisfy the given conditions yet still fail to explain the observations, and the model has no mechanism to detect or correct this mismatch. Closing the abductive cycle requires revising a discrete, structured hypothesis, which large language models cannot do reliably, even though they iterate readily in natural language. We propose HypoAgent, an agentic hypothesis refinement framework that treats condition signals not only as expressions of user intent but also as operators that steer generation. A Hypothesis Proposal Agent calls a small trained generator to propose a hypothesis. Root-cause analysis then uses fragment-level coverage to locate branches worth inspecting and gathers neighborhood evidence from the training graph. A Hypothesis Refiner Agent combines this diagnosis with previously generated hypotheses to produce updated conditions and directly revised hypotheses. HypoAgent outperforms one-shot generation in single-turn and multi-turn settings on BioKG, PharmKG8k, and DBpedia50, and in the unconditional setting on DBpedia50. Our code is available at https://github.com/HKUST-KnowComp/HypoAgent.

cs.AI↗

arXiv2026

Rethinking Prospect Theory for LLMs: Revealing the Instability of Decision-Making under Epistemic Uncertainty

Real-world decision-making often involves uncertainty expressed in linguistic rather than numerical terms, and Prospect Theory (PT) provides a classic framework for modeling human behavior under such uncertainty. Although recent studies have developed frameworks to estimate PT parameters for Large Language Models (LLMs), few have examined whether PT itself adequately describes LLM decision-making behavior. To address these gaps, we develop a streamlined workflow grounded in a classic behavioral economics experimental paradigm. First, we estimate PT parameters and evaluate how well the resulting model captures LLM decision-making behavior. We then derive probability mappings for epistemic markers in the same context and inject them into prompts to examine the stability of PT parameters under linguistic uncertainty. Our findings suggest that PT does not consistently provide a reliable account of LLM decision-making across models, and that its application to LLMs is likely sensitive to epistemic uncertainty. The findings caution against the deployment of PT-based frameworks in real-world applications where epistemic ambiguity is prevalent, giving valuable insights in behaviour interpretation and future alignment direction for LLM decision-making.

cs.AI↗

arXiv2026

NOVA: NOise-aware Verbal Confidence CAlibration for Robust Large Language Models in RAG Systems

Accurately assessing model confidence is essential for deploying large language models (LLMs) in mission-critical factual domains. While retrieval-augmented generation (RAG) is widely adopted to improve grounding, confidence calibration in RAG settings remains poorly understood. We conduct a systematic study across four benchmarks, revealing that LLMs exhibit poor calibration performance especially when noisy contexts are retrieved. Specifically, contradictory or irrelevant evidence tends to exacerbate the model's overconfidence issue. To address this, we propose NOVA Rules (NOise-Aware Verbal Confidence CAlibration Rules) to provide a principled foundation for resolving overconfidence under noise. We further design NOVA, a noise-aware calibration framework that synthesizes supervision from ~2K HotpotQA examples guided by these rules. By performing supervised fine-tuning (SFT) with this data, NOVA equips models with intrinsic noise awareness without relying on stronger teacher models. Empirical results show that NOVA yields substantial gains, improving ECE scores by 10.9% in-domain and 8.0% out-of-domain. By bridging the gap between retrieval noise and verbal calibration, NOVA paves the way for both accurate and epistemically reliable LLMs.

cs.CL↗

arXiv2026

Can LLMs Time Travel? Enhancing Temporal Consistency in Legal Agentic Search through Reinforcement Learning

While large language models (LLMs) augmented with agentic search capabilities show promise for legal reasoning, they overlook a fundamental constraint that applicable law must match the temporal context of each case, as retroactive application of statutes violates core legal principles and leads to erroneous conclusions. Our observations reveal that current legal LLMs suffer from temporal bias anchored to their training cutoff, while search agents rarely incorporate temporal constraints into queries, and that web search alone cannot provide the precise statute and precedent citations that legal reasoning demands. To address these challenges, we propose LegalSearch-R1, an end-to-end reinforcement learning framework that pairs local statute RAG for precise article matching with online web search for broader legal knowledge, trained on temporally-indexed data spanning multiple amendment periods to enforce temporal consistency. Extensive experiments on our benchmark covering 13 legal tasks demonstrate that our 7B-parameter agent outperforms state-of-the-art deep research frameworks and specialized legal LLMs by 12.9% to 29.8%, surpasses baselines by 57.7% to 80.3% on temporal consistency, and exhibits robust out-of-domain generalization. The code and data are available at https://github.com/AlexFanw/LegalSearch-R1.

cs.CL↗

arXiv2026

AgentIdeaBench: Benchmarking Scientific Ideation in the Agent Era

Scientific ideation is the capacity to formulate novel and testable hypotheses from scientific evidence, and autonomous AI scientists depend on it. Existing evaluations largely assess it by asking models to generate ideas from a static, curated set of reference papers. That passive setup departs from the retrieval-and-reasoning workflow of modern AI scientists, and it becomes less discriminative as models improve. We introduce AgentIdeaBench, a multidisciplinary benchmark that evaluates scientific ideation under two matched settings, static observation and active exploration. We report matched Static-Active evaluations for 33 LLMs across 40 densely scored subfields spanning five disciplines, using a multidimensional, literature-verified scoring framework whose critics assess originality against retrieved prior art. Active exploration reveals considerably more capability headroom, and that headroom is unevenly distributed across models. Performance scales about twice as fast as under static observation, and the exploration gain is capability-gated, favoring the strongest models over the weakest. The gain reflects better grounding, improving feasibility, clarity, and specificity while leaving measured originality unchanged under our critics. We further explore Scientific World Modeling, a generation-time loop that refines a draft hypothesis through structured thought experiments. It benefits mid-capability models, and its impact diminishes among frontier models that appear to have internalized such reasoning patterns already. AgentIdeaBench gives future work on scientific ideation a measurement basis suited to the agent era.

cs.AI↗

arXiv2026

MultivationBench: A Benchmark for Multimodal Sequential Motivation Reasoning

Multimodal Large Language Models have sparked significant interest due to their potential for social intelligence; however, their ability to perform sequential motivation reasoning remains insufficiently studied. Existing evaluations predominantly examine static text or isolated visual snapshots, which do not reflect the cumulative nature of real-world behavioral drivers. To address this gap, we introduce MultivationBench, a benchmark designed to rigorously evaluate multimodal motivation reasoning within story-driven visual narratives. The benchmark builds upon established psychological frameworks - Maslow's hierarchy and Reiss's basic desires - and requires models to integrate accumulated multimodal context to infer evolving motivations. Results indicate that MultivationBench presents a significant challenge: all tested models struggle to maintain consistent motivation reasoning across sequential contexts, revealing a critical disconnect between static recognition capabilities and the dynamic reasoning essential for human-like social understanding.

cs.AI↗

arXiv2026

Legal Rule Induction: Towards Generalizable Principle Discovery from Analogous Judicial Precedents

Legal rules encompass not only codified statutes but also implicit adjudicatory principles derived from precedents that contain discretionary norms, social morality, and policy. While computational legal research has advanced in applying established rules to cases, inducing legal rules from judicial decisions remains understudied across jurisdictions. The advent of Large Language Models (LLMs) offers unprecedented potential to automate the extraction of such latent principles, yet progress is stymied by the absence of formal task definitions and benchmarks. To address this gap, we formalize Legal Rule Induction (LRI) as the task of deriving concise, generalizable doctrinal rules from analogous precedents, distilling their shared preconditions, normative behaviors, and legal consequences. We further propose a reproducible pipeline for LRI dataset construction and, instantiating it on Chinese law as a representative jurisdiction, introduce the first LRI benchmark comprising 5,121 case sets (38,088 court cases) for model tuning and 216 expert-annotated gold test sets. Experimental results reveal that: 1) State-of-the-art LLMs struggle with over-generalization and hallucination; 2) Training on our dataset markedly enhances LLMs capabilities in capturing nuanced rule patterns across similar cases.

cs.CL↗

arXiv2026

Finding the Right Evidence: Factor-Guided Coarse-to-Fine Reasoning for Long Videos

While LVLMs rapidly improve, long-video question answering still remains challenging: relevant evidence is sparse, and question-relevant context often fails to provide cues that discriminate the correct answer from plausible alternatives. Diagnostic analysis on a manually annotated subset of MMR-V shows that prior agentic systems substantially improve cue retrieval over direct VLM inference yet fail to achieve a corresponding gain in answer accuracy, indicating that the bottleneck lies in option-discriminative evidence rather than topical relevance alone. We propose PACE (Progressive Acquisition of Critical Evidence), a factor-guided framework for long-video evidence acquisition. PACE proceeds in two stages: it first indexes clip-level descriptions guided by question-derived factors without observing the candidate answers; it then uses the candidate answers to derive contrastive cues and queries the index for verification. On MMR-V with the open-source Qwen3-VL backbone, PACE achieves 42.6% accuracy, outperforming direct inference and prior agentic baselines including Deep Video Discovery (DVD). On the same diagnostic subset, PACE recovers 66.9% of the annotated cues, providing empirical evidence that its gains are associated with improved evidence recovery rather than stronger answer-side priors alone. Consistent gains over DVD on LVBench, Video-MME, EgoSchema, and LongVideoBench suggest that option-aware evidence acquisition transfers beyond MMR-V. Code is available at https://github.com/HKUST-KnowComp/PACE.

cs.CV↗

arXiv2026

PatchWorld: Gradient-Free Optimization of Executable World Models for Agent Environments

World models for interactive text agents must typically be learned from observation-action trajectories alone. Specifically, the environment returns text observations after each action, but does not expose a ground-truth latent state nor an inspectable transition model.A research gap remains in how to induce executable code as a world model in this black-box setting for prediction and agent decision making. We introduce PatchWorld, a gradient-free framework that turns offline trajectories into executable Python world models through counterexample-guided code repair.Instead of predicting the next observation with a black-box model, PatchWorld induces symbolic belief-state programs whose action updates can be inspected, replayed, and locally patched. Across seven AgentGym environments, PatchWorld-Simple achieves the highest code-based decision-making score among evaluated methods (76.4% macro success in live one-step lookahead), matching or exceeding LLM-based lookahead while invoking no LLM calls inside the world-model prediction module itself. We further find that a human-specified residual-memory bias improves surface observation fidelity but weakens agent decision-making utility. This reveals a tradeoff in executable world models, since improving observation fidelity can come at the expense of action-discriminative dynamics, and vice versa. Code is available at https://github.com/HKBU-KnowComp/PatchWorld.

cs.CL↗