arXiv subjects

Eugene Park

Publications and source records attributed to Eugene Park.

11 recordsLinked to original sources

arXiv2026

Reflexivity as Prompt: Does Awareness of Self-Reinforcing Market Dynamics Improve LLMs as Financial Market Forecasters?

We study how frontier large language models (LLMs) behave as financial forecasters during boom-bust market cycles when made progressively aware of Soros's theory of reflexivity. Standard AI-assisted forecasting treats the market as an exogenous system. Reflexivity theory holds otherwise: prices shape fundamentals, and every forecaster is a participative agent in the loop it analyzes. We evaluate three frontier models - GPT5, Claude Sonnet 4.6, and Gemini 3 Pro - under four accumulating zero-shot conditions across two historically distinct episodes: the dot-com bubble (1996-2001) and the global financial crisis (2004-2009). The primary metric is directional forecasting accuracy; we also report the Sharpe ratio of an implied long/cash strategy to capture the risk-adjusted economic value of the forecasts. All inputs are anonymized and normalized to guard against memorization. We find that conditions incorporating reflexivity awareness improve forecasting accuracy differently across models and context windows, revealing that the same theoretical awareness can produce qualitatively different forecasting behavior across frontier LLMs.

q-fin.ST↗

arXiv2026

TeamBench: Evaluating Agent Coordination under Enforced Role Separation

Agent systems often decompose a task across multiple roles, but these roles are typically specified by prompts rather than enforced by access controls. Without enforcement, a team pass rate can mask whether agents actually coordinated or whether one role effectively did another role's work. We present TeamBench, a benchmark with 851 task templates and 931 seeded instances for evaluating agent coordination under operating system-enforced role separation. TeamBench separates specification access, workspace editing, and final certification across Planner, Executor, and Verifier roles, so that no role can read the full requirements, modify the workspace, and certify the final answer. Prompt-only and sandbox-enforced teams reach statistically indistinguishable pass rates, but prompt-only runs produce 3.6 times more cases where the verifier attempts to edit the executor's code. Verifiers approve 49% of submissions that fail the deterministic grader, and removing the verifier improves mean partial score in the ablation. Team value is also conditional. Teams benefit when single agents struggle, but hurt when single agents already perform well. A 40-session human study under the same role separation shows that our benchmark exposes interaction patterns that pass rate misses. Solo participants work through the task directly, human participants paired with agents often collapse into quick approval, and human teams spend more effort coordinating missing information across roles.

cs.AI↗

arXiv2025

AC magnetometry of van der Waals magnets using ultrasensitive Graphene Hall sensors

Probing the dynamical magnetic properties of two-dimensional (2D) materials requires sensitive techniques capable of detecting small magnetic fields from nanoscale samples. We demonstrate quantitative AC and DC magnetometry of a ferromagnetic Fe3-xGeTe2 nanoflakes using ultrasensitive graphene Hall sensors. These devices achieve record-low magnetic field detection noise at both cryogenic and room temperature, enabled by hBN encapsulation, low-resistance fluorographene contacts, and a novel fabrication process. We perform quantitative AC susceptibility measurements up to 1 kHz, resolving both real and imaginary components with nanotesla-scale sensitivity and milliradian phase accuracy, the first such measurement in a van der Waals magnet. Our results establish graphene Hall sensors as a powerful and broadly applicable platform for studying magnetic and superconducting phases near the 2D limit.

cond-mat.mtrl-sci↗

arXiv2025

InvThink: Premortem Reasoning for Safer Language Models

We present InvThink, a training and prompting framework that requires the model to enumerate, analyze, and constrain potential failures before generating its final response. Unlike existing safety alignment methods that optimize only for safe final responses, InvThink structures generation into three steps: (1) enumerate potential harms, (2) analyze their consequences, (3) generate the response under explicit mitigation constraints. We observe three findings: (i) InvThink shows higher safety scores at larger model sizes, compared to existing safety prompting and alignment baselines. (ii) InvThink mitigates the safety tax. Models trained with INVTHINK preserve their reasoning capability on standard benchmarks. (iii) beyond general safety tasks, InvThink also reduces harmful behavior in professional ethics domains (medicine, finance, law) and in agentic misalignment scenarios, achieving up to 32% reduction in harmfulness over zero-shot baselines and 16% over SafetyPrompt. We extend InvThink with supervised fine-tuning, and GRPO-based reinforcement learning across three LLM families.

cs.AI↗

arXiv2025

Tiered Agentic Oversight: A Hierarchical Multi-Agent System for Healthcare Safety

Large language models (LLMs) deployed as agents introduce significant safety risks in clinical settings due to their potential for error and single points of failure. We introduce Tiered Agentic Oversight (TAO), a hierarchical multi-agent system that enhances AI safety through layered, automated supervision. Inspired by clinical hierarchies (e.g., nurse-physician-specialist) in hospital, TAO routes tasks to specialized agents based on complexity, creating a robust safety framework through automated inter- and intra-tier communication and role-playing. Crucially, this hierarchical structure functions as an effective error-correction mechanism, absorbing up to 24% of individual agent errors before they can compound. Our experiments reveal TAO outperforms single-agent and other multi-agent systems on 4 out of 5 healthcare safety benchmarks, with up to an 8.2% improvement. Ablation studies confirm key design principles of the system: (i) its adaptive architecture is over 3% safer than static, single-tier configurations, and (ii) its lower tiers are indispensable, as their removal causes the most significant degradation in overall safety. Finally, we validated the system's synergy with human doctors in a user study where a physician, acting as the highest tier agent, provided corrective feedback that improved medical triage accuracy from 40% to 60%. Project Page: https://tiered-agentic-oversight.github.io/

cs.AI↗

arXiv2025

BehaviorSFT: Behavioral Token Conditioning for Clinical Agents Across the Proactivity Spectrum

Large Language Models (LLMs) as clinical agents require careful behavioral adaptation. While adept at reactive tasks (e.g., diagnosis reasoning), LLMs often struggle with proactive engagement, like unprompted identification of critical missing information or risks. We introduce BehaviorBench, a comprehensive dataset to evaluate agent behaviors across a clinical assistance spectrum, ranging from reactive query responses to proactive interventions (e.g., clarifying ambiguities, flagging overlooked critical data). Our BehaviorBench experiments reveal LLMs' inconsistent proactivity. To address this, we propose BehaviorSFT, a novel training strategy using behavioral tokens to explicitly condition LLMs for dynamic behavioral selection along this spectrum. BehaviorSFT boosts performance, achieving up to 97.3% overall Macro F1 on BehaviorBench and improving proactive task scores (e.g., from 95.0% to 96.5% for Qwen2.5-7B-Ins). Crucially, blind clinician evaluations confirmed BehaviorSFT-trained agents exhibit more realistic clinical behavior, striking a superior balance between helpful proactivity (e.g., timely, relevant suggestions) and necessary restraint (e.g., avoiding over-intervention) versus standard fine-tuning or explicit instructed agents.

cs.CL↗

arXiv2025

VocalAgent: Large Language Models for Vocal Health Diagnostics with Safety-Aware Evaluation

Vocal health plays a crucial role in peoples' lives, significantly impacting their communicative abilities and interactions. However, despite the global prevalence of voice disorders, many lack access to convenient diagnosis and treatment. This paper introduces VocalAgent, an audio large language model (LLM) to address these challenges through vocal health diagnosis. We leverage Qwen-Audio-Chat fine-tuned on three datasets collected in-situ from hospital patients, and present a multifaceted evaluation framework encompassing a safety assessment to mitigate diagnostic biases, cross-lingual performance analysis, and modality ablation studies. VocalAgent demonstrates superior accuracy on voice disorder classification compared to state-of-the-art baselines. Its LLM-based method offers a scalable solution for broader adoption of health diagnostics, while underscoring the importance of ethical and technical validation.

cs.SD↗

arXiv2025

Magnetically Modulated Electrical Switching in an Antiferromagnetic Transistor

A spin version of transistor, where magnetism is used to influence electrical behaviors of the semiconductor, has been a long-pursued device concept in spintronics. In this work, we experimentally study a field-effect transistor with CrSBr, a van der Waals (vdW) antiferromagnetic semiconductor, as the channel material. Unlike the weak magnetic tunability of in-plane currents previously reported in vdW magnets, the channel current of our transistor is efficiently tuned by both gate voltage and magnetic transitions, achieving a magnetoresistance ratio as high as 1500%. Combining measurement and theoretical modeling, we reveal magnetically modulated carrier concentration as the origin of the large magnetoresistance. The strategy of using both magnetic ordering and electric field in the same device to control ON/OFF states of a transistor opens a new avenue of energy-efficient spintronics for memory, logic and magnetic sensing applications.

cond-mat.mtrl-sci↗

arXiv2025

Medical Hallucinations in Foundation Models and Their Impact on Healthcare

Hallucinations in foundation models arise from autoregressive training objectives that prioritize token-likelihood optimization over epistemic accuracy, fostering overconfidence and poorly calibrated uncertainty. We define medical hallucination as any model-generated output that is factually incorrect, logically inconsistent, or unsupported by authoritative clinical evidence in ways that could alter clinical decisions. We evaluated 11 foundation models (7 general-purpose, 4 medical-specialized) across seven medical hallucination tasks spanning medical reasoning and biomedical information retrieval. General-purpose models achieved significantly higher proportions of hallucination-free responses than medical-specialized models (median: 76.6% vs 51.3%, difference = 25.2%, 95% CI: 18.7-31.3%, Mann-Whitney U = 27.0, p = 0.012, rank-biserial r = -0.64). Top-performing models such as Gemini-2.5 Pro exceeded 97% accuracy when augmented with chain-of-thought prompting (base: 87.6%), while medical-specialized models like MedGemma ranged from 28.6-61.9% despite explicit training on medical corpora. Chain-of-thought reasoning significantly reduced hallucinations in 86.4% of tested comparisons after FDR correction (q < 0.05), demonstrating that explicit reasoning traces enable self-verification and error detection. Physician audits confirmed that 64-72% of residual hallucinations stemmed from causal or temporal reasoning failures rather than knowledge gaps. A global survey of clinicians (n = 70) validated real-world impact: 91.8% had encountered medical hallucinations, and 84.7% considered them capable of causing patient harm. The underperformance of medical-specialized models despite domain training indicates that safety emerges from sophisticated reasoning capabilities and broad knowledge integration developed during large-scale pre-training, not from narrow optimization.

cs.CL↗

arXiv2024

Direct visualization of defect-controlled diffusion in van der Waals gaps

Diffusion processes govern fundamental phenomena such as phase transformations, doping, and intercalation in van der Waals (vdW) bonded materials. Here, we quantify the diffusion dynamics of W atoms by visualizing the motion of individual atoms at three different vdW interfaces: BN/vacuum, BN/BN, and BN/WSe2, by recording scanning transmission electron microscopy movies. Supported by density functional theory calculations, we infer that in all cases diffusion is governed by intermittent trapping at electron beam-generated defect sites. This leads to diffusion properties that depend strongly on the number of defects. These results suggest that diffusion and intercalation processes in vdW materials are highly tunable and sensitive to crystal quality. The demonstration of imaging, with high spatial and temporal resolution, of layers and individual atoms inside vdW heterostructures offers possibilities for direct visualization of diffusion and atomic interactions, as well as for experiments exploring atomic structures, their in-situ modification, and electrical property measurements of active devices combined with atomic resolution imaging.

cond-mat.mtrl-sci↗

arXiv2024

Electric-Field Control of Magnetic Skyrmion Chirality in a Centrosymmetric 2D van der Waals Magnet

Two-dimensional van der Waals magnets hosting topological magnetic textures, such as skyrmions, show promise for applications in spintronics and quantum computing. Electrical control of these topological spin textures would enable novel devices with enhanced performance and functionality. Here, using electron microscopy combined with in situ electric and magnetic biasing, we show that the skyrmion chirality, whether left-handed or right-handed, in insulating Cr2Ge2Te6, is controlled by external electric field direction applied during magnetic field cooling process. The electric-field-tuned chirality remains stable, even amid variations in magnetic and electric fields. Our theoretical investigation reveals that nonzero Dzyaloshinskii-Moriya interactions between the nearest neighbors, induced by the external electric field, change their sign upon reversing the electric field direction, thereby facilitating chirality selection. The electrical control of magnetic chirality demonstrated in this study can be extended to other non-metallic centrosymmetric skyrmion-hosting magnets, opening avenues for future device designs in topological spintronics and quantum computing.

cond-mat.mes-hall↗