P-Bench is built, a benchmark comprising 425 open-ended, realistic hypothesis-testing tasks spanning economics, biology, and medicine and introduces Fisher-R1, an open-weight LLM agent trained for rigorous hypothesis testing using synthetic tasks and reinforcement learning.
Abstract
Reliable hypothesis testing is the foundation of many empirical scientific claims. Large language model (LLM) agents are increasingly used to automate this process, as they can inspect datasets, generate code, and produce analyses end-to-end. However, we show that they frequently make subtle inferential errors that lead to incorrect conclusions despite correctly executed analyses. Existing benchmarks fail to capture this failure mode, as they rarely assess whether a reported p-value is statistically valid given the assumptions underlying the data. We address this gap by building P-Bench, a benchmark comprising 425 open-ended, realistic hypothesis-testing tasks spanning economics, biology, and medicine. Each task requires an agent to select a statistical method, compute a p-value, and draw a conclusion given only a scientific hypothesis and a dataset. We further introduce Fisher-R1, an open-weight LLM agent trained for rigorous hypothesis testing using synthetic tasks and reinforcement learning. On P-Bench, Fisher-R1-14B substantially improves over its backbone and outperforms strong proprietary and open-source baselines, including GPT-5.4 and DeepSeekV4-Pro, achieving a 21% average relative improvement in single-trial success over DeepSeek-V4-Pro, with gains up to 26% on the most challenging tasks. Our results demonstrate that current LLM agents lack reliable statistical reasoning for hypothesis testing and that reinforcement learning on tasks with verified statistical reward substantially improves reliability.
As large language models (LLMs) become increasingly integrated into analytical workflows, an urgent question arises: Can these models replace the trained statistician? This paper presents a controlled experiment to directly test the statistical reasoning ability of LLMs. We employ Monte Carlo simulations to generate datasets with known ground‐truth parameters and pose four canonical statistical questions to five commercially prominent models across three linguistically distinct prompt formulations and five sampling temperature settings, yielding 3000 observations in a full‐factorial design. In order to find an answer to our question, we design prompts that reflect how decision‐makers with varying degrees of statistical knowledge would query AI in the absence of a trained statistician. We find that prompts that portray higher statistical competency can result in higher accuracy for some (but not all) LLMs; we also find that LLMs can fail catastrophically on tasks requiring quantitative precision. We connect these findings to architectural differences among models and to recent literature on epistemic mirroring in LLMs and argue that the observed patterns reveal models are performing linguistic pattern matching on statistically flavoured text rather than genuine statistical reasoning. We conclude that current LLMs cannot replace the statistician, though certain architectures approach useful performance on pattern‐recognition subtasks.
Wolfgang Jank, Bernhard Klingenberg, Sonal Prabhune et al.· International Statistical Re...· 1 citation
Multimodal large language models (MLLMs) are increasingly capable scientific assistants, yet they remain far from fully autonomous research. This transition requires models to actively inspect academic papers, build global evidence views, and make traceable judgments without prespecified issues or evidence. However, existing work provides limited task paradigms or training studies for such issue- and evidence-absent verification. We study this challenge through scientific error detection, where models must determine whether errors exist and justify them with evidence-based reasoning. To fill this gap, we present VERA-RL, a reinforcement-learning formulation for scientific error detection over academic papers. Following a Reason--Verify--Scan progression, we construct VERA-13K, a 12,900-sample dataset organized into 4,300 matched chains, covering 6 scientific-error categories across the research workflow and broad natural-science domains. We further introduce fine-grained rewards for reasoning completeness, evidence alignment, and error precision. Training Qwen3-VL-8B with VERA-RL substantially improves verifiable reasoning, approaching flagship MLLMs such as Gemini 3 Pro and Qwen3-VL-235B-A22B on Scan.
Rongjin Li, Yuanxin Liu, Hao Zhou et al.· 0 citations
Although reinforcement learning with verifiable rewards (RLVR) has improved the performance of large language models (LLMs) across a variety of reasoning tasks, there is significant debate as to whether RLVR expands the reasoning capability boundary, or just improves sampling efficiency. In this paper, we investigate the nature of test-time exploration in RLVR-trained LLMs by employing controlled maze-solving experiments and extracting a tree structure from mathematical reasoning traces (BODHI-Trees) based on semantic equivalence. This helps us delineate between entropy arising from stylistic variations and genuine inferential branching. Our findings demonstrate that the policy entropy collapse observed in RLVR models is not merely syntactic, and is accompanied by a significant reduction in semantic branching entropy. While RLVR improves adherence to environmental constraints and backtracking capabilities, it constricts the space of continuations; we provide evidence suggesting that this might be responsible for the sample efficiency gains of RLVR, albeit at the cost of genuine rollout diversity.
Soumadeep Saha, Krish Sharma, Akshay Chaturvedi et al.· 1 citation
This thesis proposes a unified two-axis framework that organizes SFT and RL methods along a data axis (off-policy to on-policy) and a loss function axis (positive-only to positive-plus-negative to GRPO) and enables controlled ablations of individual components.
G. Kim, Chair Chenyan Xiong, Aditi Raghunathan· 0 citations
Large language models (LLMs) excel at answering pre-specified questions, yet their ability to navigate the open-ended, pre-conclusion stage of discovery remains largely unmeasured. We introduce Prospective Hypothesis Discovery (PHD), which asks models to autonomously construct grounded, discriminative, and testable hypothesis spaces from inconclusive evidence, including anomalous observations and fragmented records, to guide subsequent investigation. To evaluate this capability, we introduce HypoArena, comprising HypoData, a benchmark of 988 cases across six scientific and analytical domains, and HypoEval, an evaluation framework for open-ended hypothesis sets. To construct HypoData at scale, we propose Retrospective Context Regression, a Forge--Audit pipeline that reconstructs pre-conclusion contexts from completed expert documents by removing explicit conclusions, target hypotheses, and retrospective causal attributions while preserving the factual substrate. Because PHD admits multiple valid outputs, HypoEval combines bidirectional pairwise judgments with Bradley--Terry--Davidson aggregation for ranking and six-dimensional rubric scoring for diagnosis. Experiments on 15 frontier LLMs reveal clear capability stratification and model-dependent effects of structured analytical skills, with gains for several lower-performing models on HypoArena but regressions for other systems, including a top-performing model. Compared with absolute rubric scoring, arena evaluation resolves finer-grained differences among models, with aggregated rankings showing strong agreement with human experts and an independent judge. Together, these results support treating PHD as a distinct target for evaluating how LLMs formulate investigative directions when final conclusions are withheld. Our code and data are publicly available at github.com/SKYLENAGE-AI/HypoArena and github.com/SKYLENAGE-AI/HypoArena.
Tianyun Zhong, Wangyi Jiang, Wei Wang et al.· 0 citations
Large language models (LLMs) have enabled AI scientist systems to automate scientific discovery, yet existing approaches most rely on static prompting or fixed workflows and fail to accumulate experience for continual improvement. We propose HypoForge, an experience-guided multi-agent framework that learns reusable scientific skills for automated hypothesis generation and hypothesis testing. HypoForge is built on the observation that these two stages involve different supervision signals. For hypothesis generation, where explicit feedback is unavailable, HypoForge adopts an adversarial generator--discriminator mechanism to improve reasoning through comparative critique. For hypothesis testing, where empirical feedback is available, HypoForge learns testing skills from execution outcomes and ground-truth results. By matching skill learning strategies with stage-specific supervision, HypoForge enables continual improvement without fine-tuning foundation models. Experiments on hypothesis generation and testing benchmarks show that HypoForge consistently outperforms existing AI scientist frameworks and skill-level variants. Further analysis demonstrates the effectiveness of the proposed stage-specific skill learning paradigms.
Ziqing Qian, Jiaying Lei, Yi-Fang Wang et al.· 0 citations