Research
On-device research index

arXiv research

A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.

168,657 papers · 148 categories

Trend · papers per month

21416282 · May 202619922001200920172026
48 results for hallucinated trajectories

HAMBO estimates policy performance by hallucinating worst-case trajectories, providing valid lower bounds.

problem Conservative off-policy evaluation of policies in real-world applications.
method HAMBO hallucinates worst-case trajectories based on learned model uncertainty.
result Valid lower bounds on policy performance, converging to true expected return under regular conditions.

MMD-Flagger detects hallucinations in LLMs by tracking MMD between outputs and temperature-generated counterparts.

problem Detecting hallucinations in large language models.
method Maximum Mean Discrepancy (MMD) to track the difference between model outputs and temperature-generated counterparts.
result MMD-Flagger detects most hallucinations by analyzing the shape of the MMD trajectory.

Auto-CEI improves LLM reasoning by balancing assertiveness and conservativeness.

problem Hallucinations and laziness in LLM reasoning tasks.
method Expert Iteration explores reasoning trajectories, guiding incorrect paths back on track and promoting appropriate 'I don't know' responses.
result Auto-CEI achieves superior alignment in logical reasoning, mathematics, and planning tasks.

Study finds uncertainty estimators weakly correlate with LLM hallucinations.

problem Characterizing the relationship between uncertainty estimators and LLM hallucinations.
method Systematic empirical study of diverse uncertainty estimators across hallucination types and benchmarks.
result Uncertainty estimators weakly correlate with LLM hallucinations, depending on hallucination type and LLM.

Hallucinations in models are mislinked estimates, not errors.

problem Hallucinations in generative models as failures to link estimates to plausible causes.
method Formalized hallucinations, showed even optimal estimators hallucinate, provided a general lower bound on hallucinate rate, reframed hallucination as structural misalignment, and experimentally supported theory.
result Hallucinations are structural misalignments between loss minimization and human-acceptable outputs, leading to estimation errors.

A new method detects hallucinations in medical image restoration using Fourier Ring Correlation.

problem Detecting hallucinations in DL outputs for medical image restoration.
method sFRC (s(c)can FRC) over small patches of DL outputs and reference images.
result sFRC effectively detects hallucinations in CT and MRI restoration problems.

Work shows hallucination detection by LLMs is impossible without expert feedback.

problem Detecting hallucinations in LLMs is theoretically impossible without expert-labeled feedback.
method Investigated hallucination detection using a theoretical framework inspired by language identification.
result Automated hallucination detection is impossible for most language collections without expert-labeled feedback.

HalluWorld benchmarks hallucinations in language models across diverse tasks.

problem Inconsistent benchmarks of hallucination in language models.
method Explicit reference-world formulation, synthetic and semi-synthetic environments, controlled variation of world complexity.
result Models struggle with perceptual hallucination but not multi-step state tracking.

New framework assesses AI hallucinations in inverse problems.

problem Artificial intelligence can produce incorrect details in imaging problems.
method Theoretical framework and algorithms to estimate and assess hallucinations.
result Developed necessary and sufficient conditions for hallucinations and computable bounds.

ECLIPSE detects AI hallucinations in finance with high accuracy.

problem Hallucinations in AI-generated answers limit safe deployment in finance.
method Combines entropy estimation and perplexity decomposition to measure model evidence use.
result ECLIPSE achieves ROC AUC of 0.89 and average precision of 0.90 on financial QA dataset.

Graph signal processing detects hallucinations in large language models.

problem Detecting factual reasoning from hallucinations in large language models.
method Modeling transformer layers as dynamic graphs, using spectral analysis to define diagnostics.
result Spectral signatures can distinguish different types of hallucinations and achieve high accuracy.

The paper explores fundamental limits of learning non-hallucinating generative models.

problem Hallucinations in generative models producing invalid outputs.
method Developed a theoretical framework to analyze learnability from a learning-theoretic perspective, incorporating inductive biases.
result Non-hallucinating learning is statistically impossible without additional inductive biases.

Paper proposes a method to reduce hallucinations in diffusion models using Laplacian score sharpening.

problem Hallucinations in diffusion models create incoherent or unrealistic samples.
method Post-hoc adjustment to the score function during inference using Laplacian approximation.
result Significantly reduces the rate of hallucinated samples across various data types.

Paper estimates AI hallucinations in conditional generation tasks.

problem Estimating the frequency of AI-generated incorrect responses.
method Developed a method to estimate hallucination probability from generated responses and log probabilities.
result Method accurately estimates hallucination rate in natural language and synthetic tasks.

Benchmark evaluates AI-generated financial QA hallucinations, highlighting system vulnerabilities.

problem Ensuring factual accuracy of AI-generated financial QA outputs.
method Developed a benchmark dataset and evaluated six detection methods under clean and noisy conditions.
result LLM-based judges and embedding methods perform best, but degrade under noisy conditions.

Paper proves impossible for large language models to control hallucinations without sacrificing other properties.

problem Achieving truthful knowledge representation, semantic information conservation, and knowledge-constrained optimality simultaneously in large language models.
method Modeling inference as an auction of ideas, using mechanism design, proper scoring rules, and transformer architecture analysis.
result No LLM can simultaneously achieve all four essential properties without violating at least one.

Paper introduces SDM for detecting LLM hallucinations, improving on entropy tests.

problem Challenges of Large Language Models (LLMs) with non-factual, nonsensical responses.
method Joint clustering on sentence embeddings to measure semantic divergence between prompts and responses.
result SDM framework detects deeper form of arbitrariness in LLM responses.

The paper addresses Dyna-style RL's value hallucination issue by proposing a new algorithm.

problem Value hallucination in Dyna-style RL due to bootstrapping simulated states.
method Introduces a new Dyna algorithm using predecessor models with multi-step updates.
result Evidence supports the Hallucinated Value Hypothesis (HVH), suggesting predecessor models with multi-step updates are promising.

Study compares DSPy teleprompter algorithms for aligning LLM evaluations with human annotations.

problem Aligning LLM evaluation metrics with human annotations.
method Comparative analysis of five teleprompter algorithms within the DSPy framework.
result Certain teleprompters outperform others in detecting hallucinations.

New findings show language models can't simultaneously avoid hallucinations and capture all language richness.

problem Achieving both valid output and full language richness in language generation.
method Investigates language generation within a statistical setting, focusing on consistency and breadth.
result For most collections of candidate languages, a language model cannot simultaneously avoid hallucinations and capture all language richness.

We characterize language generation with stability and breadth, proving impossibility results.

problem Characterizing and proving impossibility results for language generation with stability and breadth.
method Analysis of existing notions of breadth and stability, proving lower bounds.
result Proven impossibility of generating with higher perplexity or lower hallucination rate for stable generators.

New method uses LLMs to extract financial insights from Q&A sections of reports.

problem Scalability and accuracy issues in extracting valuable insights from financial report Q&A sections.
method Combines retrieval-augmented generation technique with metadata.
result Empirically demonstrates superior performance of the proposed method.

Generative Score Inference improves uncertainty quantification for multimodal data.

problem Accurate uncertainty quantification in multimodal learning tasks.
method Generative Score Inference (GSI) uses synthetic samples to approximate conditional score distributions.
result GSI achieves state-of-the-art performance in hallucination detection and image captioning uncertainty estimation.

New framework tackles deep financial reporting bottleneck by improving hallucination and coherence.

problem Statistical smoothing trap in LLMs limits deep financial reporting quality.
method DeepNews Framework integrates information foraging, schema-guided planning, and adversarial prompting.
result DeepNews system achieves 25% acceptance rate in blind test, significantly outperforming SOTA.

Financial institutions face new model risks with AI, requiring enhanced model risk management.

problem New model risks from Generative AI applications in financial institutions.
method Enhanced model risk framework with additional testing and controls.
result Financial institutions need to enhance their model risk management for Generative AI applications.

FactTest assesses LLM factuality with Type I error control.

problem Lack of rigorous factuality verification for LLMs.
method Formulates factuality testing as hypothesis testing, ensuring Type I and II error control.
result Improves model accuracy by over 40% in abstaining from unknown questions.

Proposes H-UCRL for efficient model-based RL with sublinear regret.

problem Greedy policy exploration in model-based RL ignores epistemic uncertainty.
method Reparameterizes plausible models, hallucinates control, augments input space, solves with greedy planners.
result H-UCRL achieves provably sublinear regret for Gaussian Process models.

Transformer architecture struggles with complex tasks due to limitations in function composition.

problem Transformer architecture's limitations in handling complex tasks.
method Used Communication Complexity to prove limitations in composing functions.
result Transformer layer is incapable of handling large domain functions, even when domains are small.

Language models fail to execute simple steps, showing gating and binding errors.

problem Procedural hallucinations in language models, failing to execute simple steps.
method Analyzed long-context binding tasks, identifying gating and binding errors.
result Procedural errors are due to gating and binding failures, with recency bias contributing to the latter.

Framework controls uncertainty in LLMs without labels or probabilities.

problem Managing uncertainty in black-box LLMs without token-level probability or true labels.
method Integrates generative models, UCP, and conformal alignment to control uncertainty.
result Achieves close-to-nominal coverage and tighter thresholds than split UCP.

DS-CP improves reliability of uncertainty quantification for large language models under domain shift.

problem Overconfident and factually incorrect outputs (hallucinations) from large language models.
method Adapts conformal prediction to large language models under domain shift by reweighting calibration samples.
result DS-CP delivers more reliable coverage than standard conformal prediction, especially under substantial distribution shifts.

Language models fail to process hallucinated responses, and this study diagnoses the failure.

problem Language models fail to process hallucinated responses, leading to over-concentration or diffuse attention.
method The study uses forced scoring of benchmark-labeled responses to compute attention shapes and analyze the symmetric component of the degree-normalized attention operator.
result The study proves that every transpose-invariant spectral diagnostic of the attention operator is orientation-blind and bounds the sensitivity of any Lipschitz diagnostic by the asymmetry coefficient \(G\).

ChatGPT selects stocks for investment portfolios, but optimization models improve results.

problem Using AI for investment advice due to model inaccuracies.
method Used ChatGPT to generate a stock universe, then compared various portfolio optimization strategies.
result Combining AI-generated stock selection with advanced optimization models yields better investment outcomes.

FinanceBench benchmarks LLMs on financial QA, revealing limitations.

problem Evaluating LLMs' performance on financial question answering.
method Developed a comprehensive test suite (FinanceBench) with 10,231 questions, tested 16 models, and manually reviewed answers.
result Existing LLMs have significant limitations for financial QA, especially GPT-4-Turbo.

Transformers for binary decisions are sensitive to evidence order, leading to unreliable outcomes.

problem Order sensitivity in Transformers for binary decisions leads to unreliable outcomes.
method Formalized an expectation-realization gap and developed QMV and EDFL bounds.
result Uniform permutation mixtures reduce dispersion and improve reliability.

Two new metrics assess LLM faithfulness and entropy, improving model reliability.

problem Evaluating the accuracy of LLMs in generating coherent responses.
method Proposes SF and SEP metrics based on information theory and thermodynamics.
result High SF and SEP scores indicate more faithful LLM responses.

New method identifies shared topics in LLM inputs and outputs for better detection of hallucinations.

problem Detecting semantic drift in LLM responses from context.
method Transformed Deterministic Information Bottleneck (DIB) into UDIB for high-dimensional data.
result UDIB generates more informative topic representations for SDM, improving hallucination detection.