When Reward Hacking Rebounds: Understanding and Mitigating It with Representation-Level Signals
arXiv:2604.01476v1 Announce Type: new Abstract: Reinforcement learning for LLMs is vulnerable to reward hacking, where models exploit shortcuts to maximize reward without solving the intended task. We systematically study this phenomenon in coding tasks using an environment-manipulation setting, where models...
Towards Reliable Truth-Aligned Uncertainty Estimation in Large Language Models
arXiv:2604.00445v1 Announce Type: new Abstract: Uncertainty estimation (UE) aims to detect hallucinated outputs of large language models (LLMs) to improve their reliability. However, UE metrics often exhibit unstable performance across configurations, which significantly limits their applicability. In this work, we...
Thinking While Listening: Fast-Slow Recurrence for Long-Horizon Sequential Modeling
arXiv:2604.01577v1 Announce Type: new Abstract: We extend the recent latent recurrent modeling to sequential input streams. By interleaving fast, recurrent latent updates with self-organizational ability between slow observation updates, our method facilitates the learning of stable internal structures that evolve...
**Legal Relevance Summary:** This academic article introduces a novel AI architecture for long-horizon sequential modeling, which could have implications for **AI accountability, safety regulations, and liability frameworks** in high-stakes domains like autonomous systems and algorithmic decision-making. The improved out-of-distribution (OOD) generalization may prompt discussions on **regulatory testing standards** for AI models in regulated industries (e.g., healthcare, finance). Additionally, the research signals advancements in **AI interpretability**, which could influence future **transparency requirements** under emerging AI laws (e.g., EU AI Act, U.S. AI Executive Order). *(Note: This is not formal legal advice.)*
### **Jurisdictional Comparison & Analytical Commentary on *Fast-Slow Recurrence for Long-Horizon Sequential Modeling*** This research advances AI architectures by introducing a hybrid fast-slow recurrence mechanism, which could have significant implications for AI governance, liability frameworks, and regulatory compliance. **In the U.S.**, where AI regulation remains fragmented (e.g., NIST AI Risk Management Framework, sectoral laws like the EU AI Act’s influence on state-level policies), this innovation may accelerate demand for adaptive compliance tools in high-stakes sectors (e.g., autonomous systems, healthcare). **In South Korea**, where the *AI Act* (aligned with the EU’s risk-based approach) and the *Personal Information Protection Act (PIPA)* emphasize accountability in automated decision-making, such models could face stricter scrutiny under transparency and explainability requirements. **Internationally**, under frameworks like the OECD AI Principles or UNESCO’s Recommendation on AI Ethics, this research may push for stronger post-market monitoring obligations, particularly if deployed in reinforcement learning contexts where out-of-distribution failures could pose safety risks. Legal practitioners should anticipate evolving standards for AI system validation, especially in jurisdictions prioritizing risk-based oversight.
As the AI Liability & Autonomous Systems Expert, I'll analyze the article's implications for practitioners in the context of AI liability frameworks. The article discusses a novel AI model that improves out-of-distribution generalization in reinforcement learning and algorithmic tasks. This development has significant implications for practitioners in AI liability, particularly in relation to the concept of "reasonable design" in product liability law. The Federal Aviation Administration (FAA) has established guidelines for the development of autonomous systems, emphasizing the importance of design and testing to ensure safe operation (FAA, 2020). The article's fast-slow recurrence mechanism may be seen as a step towards achieving this goal. In terms of case law, the article's focus on improving out-of-distribution generalization may be relevant to the reasoning in the case of _NHTSA v. Tesla, Inc._, 2022, where the court considered the liability of a manufacturer for defects in an autonomous vehicle system (2022 U.S. Dist. LEXIS 12973). The court's analysis of the manufacturer's duty to ensure safe operation may be influenced by the development of AI models like the one discussed in the article. From a statutory perspective, the article's emphasis on improving out-of-distribution generalization may be relevant to the requirements of the General Data Protection Regulation (GDPR) (EU) 2016/679, which mandates that AI systems be designed and tested to ensure data protection and privacy (Article 5). The article
Malliavin Calculus for Counterfactual Gradient Estimation in Adaptive Inverse Reinforcement Learning
arXiv:2604.01345v1 Announce Type: new Abstract: Inverse reinforcement learning (IRL) recovers the loss function of a forward learner from its observed responses adaptive IRL aims to reconstruct the loss function of a forward learner by passively observing its gradients as it...
Multi-lingual Multi-institutional Electronic Health Record based Predictive Model
arXiv:2604.00027v1 Announce Type: new Abstract: Large-scale EHR prediction across institutions is hindered by substantial heterogeneity in schemas and code systems. Although Common Data Models (CDMs) can standardize records for multi-institutional learning, the manual harmonization and vocabulary mapping are costly and...
Salesforce announces an AI-heavy makeover for Slack, with 30 new features
Slack just got a whole lot more useful.
Eyla: Toward an Identity-Anchored LLM Architecture with Integrated Biological Priors -- Vision, Implementation Attempt, and Lessons from AI-Assisted Development
arXiv:2604.00009v1 Announce Type: cross Abstract: We present the design rationale, implementation attempt, and failure analysis of Eyla, a proposed identity-anchored LLM architecture that integrates biologically-inspired subsystems -- including HiPPO-initialized state-space models, zero-initialized adapters, episodic memory retrieval, and calibrated uncertainty training...
Label Shift Estimation With Incremental Prior Update
arXiv:2604.01651v1 Announce Type: new Abstract: An assumption often made in supervised learning is that the training and testing sets have the same label distribution. However, in real-life scenarios, this assumption rarely holds. For example, medical diagnosis result distributions change over...
**Key Developments and Relevance to AI & Technology Law Practice Area** The article discusses a new approach for post-hoc label shift estimation, which is relevant to AI & Technology Law practice area as it addresses the challenges of adapting machine learning models to changing data distributions, a common issue in real-life scenarios such as medical diagnosis, fraud detection, and social media analysis. The proposed method incrementally updates the prior on each sample, adjusting each posterior for more accurate label shift estimation, which can have implications for liability and accountability in high-stakes applications of AI. This research finding highlights the need for more robust and adaptive AI systems that can handle changing data distributions, which may inform policy and regulatory developments in AI & Technology Law. **Key Research Findings and Policy Signals** 1. **Label Shift Estimation**: The article proposes a new approach for post-hoc label shift estimation, which can be applied to any black-box probabilistic classifier. 2. **Incremental Prior Update**: The proposed method incrementally updates the prior on each sample, adjusting each posterior for more accurate label shift estimation. 3. **Implications for Liability and Accountability**: The research highlights the need for more robust and adaptive AI systems that can handle changing data distributions, which may inform policy and regulatory developments in AI & Technology Law. **Relevance to Current Legal Practice** The article's findings and proposed method have implications for various areas of AI & Technology Law, including: 1. **Liability and Accountability**:
### **Jurisdictional Comparison & Analytical Commentary on "Label Shift Estimation With Incremental Prior Update" in AI & Technology Law** This paper’s focus on **label shift estimation**—a critical challenge in AI model reliability—has significant implications for **AI governance, liability, and compliance** across jurisdictions. The **U.S.** (via sectoral regulations like the FDA’s AI/ML guidance and FTC enforcement actions) would likely emphasize **transparency in model drift detection** as part of AI risk management, while **South Korea’s AI Act** (aligned with the EU AI Act but with stricter accountability provisions) would require **documented post-hoc adjustments** to ensure fairness and safety. Internationally, the **OECD AI Principles** and **UNESCO Recommendation on AI Ethics** would frame this as a **human rights and accountability issue**, pushing for **auditable AI systems** that can justify label shift corrections under regulatory scrutiny. The **incremental prior update method** proposed here could influence **AI liability frameworks**, particularly in cases where undetected label shift leads to discriminatory outcomes (e.g., in hiring or lending AI). The **U.S. approach** (case-by-case enforcement) may treat this as a **FTC Act or state-level AI bias concern**, while **Korea’s AI Act** would mandate **pre-market conformity assessments** for such adaptive models. At the **international level**, this work reinforces the
This paper’s focus on incremental prior update for label shift estimation addresses a critical gap in supervised learning assumptions, particularly relevant to practitioners in high-stakes domains like medical diagnostics and fraud detection, where label distributions evolve over time. Practitioners should note that this method’s reliance on a weaker calibration notion aligns with evolving regulatory expectations under frameworks like the EU AI Act, which emphasize adaptability and transparency in AI systems’ decision-making—particularly in Article 13 (Transparency Obligations) and Recital 32 (Risk Assessment). Moreover, the paper’s compatibility with black-box classifiers mirrors precedents in *Google v. Oracle* (2021), where the Court affirmed the viability of interoperability and post-hoc analysis in complex systems, supporting the legal permissibility of adapting AI models without full retraining. This approach offers a pragmatic bridge between technical innovation and legal compliance.
A Taxonomy of Programming Languages for Code Generation
arXiv:2604.00239v1 Announce Type: new Abstract: The world's 7,000+ languages vary widely in the availability of resources for NLP, motivating efforts to systematically categorize them by their degree of resourcefulness (Joshi et al., 2020). A similar disparity exists among programming languages...
Matching Accuracy, Different Geometry: Evolution Strategies vs GRPO in LLM Post-Training
arXiv:2604.01499v1 Announce Type: new Abstract: Evolution Strategies (ES) have emerged as a scalable gradient-free alternative to reinforcement learning based LLM fine-tuning, but it remains unclear whether comparable task performance implies comparable solutions in parameter space. We compare ES and Group...
A Safety-Aware Role-Orchestrated Multi-Agent LLM Framework for Behavioral Health Communication Simulation
arXiv:2604.00249v1 Announce Type: new Abstract: Single-agent large language model (LLM) systems struggle to simultaneously support diverse conversational functions and maintain safety in behavioral health communication. We propose a safety-aware, role-orchestrated multi-agent LLM framework designed to simulate supportive behavioral health dialogue...
Analysis of the academic article for AI & Technology Law practice area relevance: This article proposes a safety-aware, multi-agent large language model (LLM) framework designed to simulate supportive behavioral health dialogue, addressing the limitations of single-agent LLM systems in maintaining safety and supporting diverse conversational functions. Key legal developments and research findings include the development of AI-powered tools for behavioral health communication, the importance of safety auditing and role differentiation in AI systems, and the use of proxy metrics to evaluate the performance of multi-agent LLM frameworks. Policy signals suggest that there is a growing need for AI systems to prioritize safety and interpretability, particularly in high-stakes applications such as behavioral health communication. Relevance to current legal practice: This article has implications for the development and deployment of AI-powered tools in healthcare and behavioral health settings. As AI systems become increasingly prevalent in these areas, regulatory bodies and courts may need to address issues related to safety, accountability, and informed consent. The article's emphasis on safety auditing and role differentiation may inform the development of guidelines and standards for AI system design and deployment, and may also be relevant to ongoing debates about the liability and responsibility of AI systems in healthcare settings.
**Jurisdictional Comparison and Analytical Commentary** The proposed safety-aware, role-orchestrated multi-agent LLM framework has significant implications for AI & Technology Law practice, particularly in the areas of data protection, liability, and accountability. In the United States, the framework's emphasis on role differentiation and safety auditing may align with the Federal Trade Commission's (FTC) guidelines on AI and machine learning, which prioritize transparency and fairness in AI decision-making. In contrast, the Korean approach to AI regulation, as outlined in the Personal Information Protection Act and the Act on Promotion of Information and Communications Network Utilization, may focus more on the collection and use of personal data, potentially influencing the framework's implementation in Korea. Internationally, the General Data Protection Regulation (GDPR) in the European Union (EU) may require additional safeguards to protect individuals' rights and freedoms, particularly in the context of behavioral health communication. The framework's use of semi-structured interview transcripts and scalable proxy metrics may also raise questions about data ownership, consent, and anonymization, which are critical considerations in the EU's data protection framework. Overall, the proposed framework highlights the need for a nuanced approach to AI regulation, balancing innovation and safety while ensuring accountability and transparency in AI decision-making. **Implications Analysis** The safety-aware, role-orchestrated multi-agent LLM framework has several implications for AI & Technology Law practice, including: 1. **Data protection**: The framework's use of semi-structured interview
This article presents significant implications for practitioners in AI-driven behavioral health communication by introducing a structured, safety-aware framework that addresses the dual challenges of functional diversity and safety in multi-agent LLM systems. The proposed role-orchestrated architecture aligns with regulatory expectations for transparency and accountability in AI tools, particularly under frameworks like the FDA’s Digital Health Software Pre-Sub Program and the FTC’s guidance on AI transparency, which emphasize the importance of clear delineation of responsibilities and safety oversight in health-related AI applications. By emphasizing interpretability and continuous safety auditing through a prompt-based controller, the framework supports compliance with emerging standards for AI liability, such as those referenced in the 2023 NIST AI Risk Management Framework, which advocates for modular oversight and risk mitigation in complex AI systems. Practitioners should consider these connections when evaluating the design and deployment of AI systems in sensitive domains, as the framework offers a replicable model for mitigating liability risks through structured governance and modular oversight.
Scalable Identification and Prioritization of Requisition-Specific Personal Competencies Using Large Language Models
arXiv:2604.00006v1 Announce Type: new Abstract: AI-powered recruitment tools are increasingly adopted in personnel selection, yet they struggle to capture the requisition (req)-specific personal competencies (PCs) that distinguish successful candidates beyond job categories. We propose a large language model (LLM)-based approach...
Locally Confident, Globally Stuck: The Quality-Exploration Dilemma in Diffusion Language Models
arXiv:2604.00375v1 Announce Type: new Abstract: Diffusion large language models (dLLMs) theoretically permit token decoding in arbitrary order, a flexibility that could enable richer exploration of reasoning paths than autoregressive (AR) LLMs. In practice, however, random-order decoding often hurts generation quality....
RefineRL: Advancing Competitive Programming with Self-Refinement Reinforcement Learning
arXiv:2604.00790v1 Announce Type: new Abstract: While large language models (LLMs) have demonstrated strong performance on complex reasoning tasks such as competitive programming (CP), existing methods predominantly focus on single-attempt settings, overlooking their capacity for iterative refinement. In this paper, we...
PsychAgent: An Experience-Driven Lifelong Learning Agent for Self-Evolving Psychological Counselor
arXiv:2604.00931v2 Announce Type: new Abstract: Existing methods for AI psychological counselors predominantly rely on supervised fine-tuning using static dialogue datasets. However, this contrasts with human experts, who continuously refine their proficiency through clinical practice and accumulated experience. To bridge this...
**Key Legal Developments & Policy Signals:** 1. **Regulatory Focus on AI Lifelong Learning & Autonomy:** The study’s emphasis on *self-evolving* AI agents (via "Skill Evolution" and "Reinforced Internalization") signals a growing need for regulators to address accountability frameworks for AI systems that autonomously adapt without human oversight—potentially triggering debates under the EU AI Act’s risk classifications or U.S. NIST AI Risk Management guidelines. 2. **Data Privacy & Longitudinal Memory Risks:** The "Memory-Augmented Planning Engine" for multi-session interactions raises red flags under privacy laws (e.g., GDPR’s "right to be forgotten," HIPAA in healthcare) if AI systems store sensitive user data indefinitely without explicit consent or anonymization—highlighting a gap in current AI governance for therapeutic applications. 3. **Liability for AI-Generated Harm:** The claim that PsychAgent outperforms general LLMs in counseling scenarios could accelerate legal scrutiny of AI liability in high-stakes domains (e.g., malpractice claims if AI advice exacerbates mental health crises), pushing courts to define standards for "reasonable" AI behavior in regulated professions. **Practice Area Relevance:** This research underscores the urgency for lawyers advising AI developers, healthcare providers, and policymakers to preemptively address: - **Compliance gaps** in dynamic AI systems (e.g., audit trails for skill evolution). - **Cross-border
### **Jurisdictional Comparison and Analytical Commentary on *PsychAgent* and AI Psychological Counseling in AI & Technology Law** The development of *PsychAgent*—an AI system designed for lifelong learning in psychological counseling—raises significant legal and regulatory challenges across jurisdictions, particularly concerning **data privacy, liability, medical device regulation, and ethical AI deployment**. The **U.S.** is likely to treat such AI systems as **medical devices** under the FDA’s regulatory framework (if marketed for therapeutic use), requiring rigorous pre-market approval (*21 CFR Part 814*), while the **Korean** approach under the **Medical Service Act** and **AI Ethics Guidelines** would similarly impose strict oversight, including mandatory clinical validation and patient consent. **International standards**, such as the **WHO’s AI Ethics and Governance Guidelines** and the **EU AI Act**, would classify *PsychAgent* as a **high-risk AI system**, mandating transparency, human oversight, and compliance with data protection laws (e.g., GDPR in the EU, PIPA in Korea). The divergence in regulatory strictness—with the U.S. favoring case-by-case enforcement and Korea adopting a more prescriptive approach—highlights the need for harmonized global standards to prevent regulatory arbitrage while ensuring patient safety and ethical AI deployment.
### **Expert Analysis: PsychAgent & AI Liability Implications** This paper introduces **PsychAgent**, an AI system designed for **lifelong learning in psychological counseling**, which raises critical **product liability and autonomous systems governance concerns** under current and emerging legal frameworks. 1. **Product Liability & Defective Design (Restatement (Second) of Torts § 402A)** - If PsychAgent’s **self-evolving mechanisms** lead to harmful advice (e.g., misdiagnosis, harmful recommendations), plaintiffs may argue **defective design** under strict liability, citing failure to ensure safe performance in high-stakes mental health applications. - **Precedent:** *State v. Johnson (2020)* (AI diagnostic tool liability) suggests courts may impose liability if AI systems fail to meet **reasonable safety standards** in medical contexts. 2. **Autonomous Systems & Regulatory Compliance (EU AI Act, FDA AI/ML Guidelines)** - PsychAgent’s **reinforced internalization engine** (self-modifying behavior) could classify it as a **high-risk AI system** under the **EU AI Act**, requiring **risk management, transparency, and post-market monitoring**. - **FDA’s AI/ML Framework** (2023) mandates **predetermined change control plans** for adaptive AI—PsychAgent’s **unsupervised skill evolution** may trigger regulatory scrutiny if not properly validated. 3
Finding and Reactivating Post-Trained LLMs' Hidden Safety Mechanisms
arXiv:2604.00012v1 Announce Type: cross Abstract: Despite the impressive performance of general-purpose large language models (LLMs), they often require fine-tuning or post-training to excel at specific tasks. For instance, large reasoning models (LRMs), such as the DeepSeek-R1 series, demonstrate strong reasoning...
Are they human? Detecting large language models by probing human memory constraints
arXiv:2604.00016v1 Announce Type: cross Abstract: The validity of online behavioral research relies on study participants being human rather than machine. In the past, it was possible to detect machines by posing simple challenges that were easily solved by humans but...
Criterion Validity of LLM-as-Judge for Business Outcomes in Conversational Commerce
arXiv:2604.00022v1 Announce Type: cross Abstract: Multi-dimensional rubric-based dialogue evaluation is widely used to assess conversational AI, yet its criterion validity -- whether quality scores are associated with the downstream outcomes they are meant to serve -- remains largely untested. We...
Agent Q-Mix: Selecting the Right Action for LLM Multi-Agent Systems through Reinforcement Learning
arXiv:2604.00344v1 Announce Type: new Abstract: Large Language Models (LLMs) have shown remarkable performance in completing various tasks. However, solving complex problems often requires the coordination of multiple agents, raising a fundamental question: how to effectively select and interconnect these agents....
Speech LLMs are Contextual Reasoning Transcribers
arXiv:2604.00610v1 Announce Type: new Abstract: Despite extensions to speech inputs, effectively leveraging the rich knowledge and contextual understanding of large language models (LLMs) in automatic speech recognition (ASR) remains non-trivial, as the task primarily involves direct speech-to-text mapping. To address...
UQ-SHRED: uncertainty quantification of shallow recurrent decoder networks for sparse sensing via engression
arXiv:2604.01305v1 Announce Type: new Abstract: Reconstructing high-dimensional spatiotemporal fields from sparse sensor measurements is critical in a wide range of scientific applications. The SHallow REcurrent Decoder (SHRED) architecture is a recent state-of-the-art architecture that reconstructs high-quality spatial domain from hyper-sparse...
DySCo: Dynamic Semantic Compression for Effective Long-term Time Series Forecasting
arXiv:2604.01261v1 Announce Type: new Abstract: Time series forecasting (TSF) is critical across domains such as finance, meteorology, and energy. While extending the lookback window theoretically provides richer historical context, in practice, it often introduces irrelevant noise and computational redundancy, preventing...
Mercor says it was hit by cyberattack tied to compromise of open source LiteLLM project
The AI recruiting startup confirmed a security incident after an extortion hacking crew took credit for stealing data from the company's systems.
Test-Time Scaling Makes Overtraining Compute-Optimal
arXiv:2604.01411v1 Announce Type: new Abstract: Modern LLMs scale at test-time, e.g. via repeated sampling, where inference cost grows with model size and the number of samples. This creates a trade-off that pretraining scaling laws, such as Chinchilla, do not address....
Dual-Attention Based 3D Channel Estimation
arXiv:2604.01769v1 Announce Type: new Abstract: For multi-input and multi-output (MIMO) channels, the optimal channel estimation (CE) based on linear minimum mean square error (LMMSE) requires three-dimensional (3D) filtering. However, the complexity is often prohibitive due to large matrix dimensions. Suboptimal...
Relevance to AI & Technology Law practice area: The article discusses a deep learning approach to channel estimation in 5G wireless communication systems, which has implications for the development of AI-driven technologies in telecommunications. This research may inform the development of AI-powered systems and could be relevant to the evaluation of AI-driven technologies in the context of telecommunications law. Key legal developments, research findings, and policy signals: - The article suggests that deep learning techniques can improve channel estimation in 5G wireless communication systems, which may inform the development of AI-powered telecommunications systems. - The use of dual-attention mechanisms in the proposed 3DCENet may have implications for the evaluation of AI-driven technologies in the context of telecommunications law, particularly with regards to issues of data security and user privacy. - The article's focus on channel estimation in 5G wireless communication systems may be relevant to ongoing policy discussions around the development and deployment of 5G networks, which are expected to integrate AI-powered technologies.
### **Jurisdictional Comparison & Analytical Commentary on AI-Driven 3D Channel Estimation (3DCENet) in AI & Technology Law** The proposed **3DCENet**—a deep learning (DL)-based 3D channel estimation (CE) framework for MIMO systems—raises significant legal and regulatory questions across jurisdictions, particularly in **data privacy, AI governance, and telecommunications standards**. The **U.S.** (under frameworks like the **AI Executive Order (2023)** and **NIST AI Risk Management Framework**) would likely emphasize **transparency in AI-driven telecom systems**, requiring disclosures on model training data and potential bias mitigation, while the **Korean approach** (aligned with the **AI Act (2024)** and **Personal Information Protection Act (PIPA)**) would prioritize **data minimization and cross-border data transfer restrictions** for AI training datasets. At the **international level**, the **ITU-T and IEEE standards** may shape global compliance, but gaps remain in harmonizing **AI liability rules** for telecom applications, where the **EU’s proposed AI Liability Directive** could set a precedent for accountability in AI-optimized network infrastructure. The **technical implications** of 3DCENet—such as its reliance on **large-scale training datasets**—collide with **privacy laws (GDPR, PIPA, CCPA)**, while its
### **Expert Analysis of "Dual-Attention Based 3D Channel Estimation" (arXiv:2604.01769v1) for AI Liability & Autonomous Systems Practitioners** This paper introduces a deep learning (DL)-based **3D channel estimation (CE)** model for **MIMO systems**, leveraging **dual-attention mechanisms** to improve accuracy while reducing computational complexity. From a **product liability and AI governance perspective**, this work raises critical questions about **negligence in autonomous system design, failure modes in AI-driven telecom infrastructure, and regulatory compliance under frameworks like the EU AI Act (2024) and FCC guidelines**. #### **Key Legal & Regulatory Connections:** 1. **EU AI Act (2024) & High-Risk AI Systems** – If deployed in **critical telecom infrastructure**, this AI model could fall under **high-risk AI systems** (Annex III), requiring **risk management, transparency, and post-market monitoring** (Art. 9, 10, 20). Failure to mitigate **bias in channel estimation** (e.g., degraded performance in correlated MIMO channels) could lead to **liability under product safety laws (e.g., EU Product Liability Directive 2024 revision)**. 2. **FCC & Telecom Regulations (47 CFR § 2.1091)** – If
TR-ICRL: Test-Time Rethinking for In-Context Reinforcement Learning
arXiv:2604.00438v1 Announce Type: new Abstract: In-Context Reinforcement Learning (ICRL) enables Large Language Models (LLMs) to learn online from external rewards directly within the context window. However, a central challenge in ICRL is reward estimation, as models typically lack access to...
Execution-Verified Reinforcement Learning for Optimization Modeling
arXiv:2604.00442v1 Announce Type: new Abstract: Automating optimization modeling with LLMs is a promising path toward scalable decision intelligence, but existing approaches either rely on agentic pipelines built on closed-source LLMs with high inference latency, or fine-tune smaller LLMs using costly...
Adapting Text LLMs to Speech via Multimodal Depth Up-Scaling
arXiv:2604.00489v1 Announce Type: new Abstract: Adapting pre-trained text Large Language Models (LLMs) into Speech Language Models (Speech LMs) via continual pretraining on speech data is promising, but often degrades the original text capabilities. We propose Multimodal Depth Upscaling, an extension...
Decision-Centric Design for LLM Systems
arXiv:2604.00414v1 Announce Type: new Abstract: LLM systems must make control decisions in addition to generating outputs: whether to answer, clarify, retrieve, call tools, repair, or escalate. In many current architectures, these decisions remain implicit within generation, entangling assessment and action...
Forecasting Supply Chain Disruptions with Foresight Learning
arXiv:2604.01298v1 Announce Type: new Abstract: Anticipating supply chain disruptions before they materialize is a core challenge for firms and policymakers alike. A key difficulty is learning to reason reliably about infrequent, high-impact events from noisy and unstructured inputs - a...