LLM agents increasingly operate through multi-turn tool use and environment interaction, where safety risks often emerge from intermediate steps long before they surface in the final outcome. Reactive auditing is therefore insufficient: post-hoc diagnosis frequently misses the chance to flag risks while they are unfolding. We propose TRACES, a representation-based proactive auditor that learns prefix-level trajectory risk states from the hidden representations of an observer LLM. TRACES induces latent mechanism features from step representations and models their temporal evolution to estimate whether a partial trajectory is drifting toward unsafe behavior. To sidestep the cost and ambiguity of step-level risk annotation, TRACES is trained with weak trajectory-level supervision while still producing dense prefix-level risk estimates. Across multiple agent safety benchmarks, TRACES improves both full-trajectory safety prediction and proactive risk discrimination. Our analyses further suggest that these risk states can help train a safer agent, highlighting the broader potential of proactive auditing for long-horizon agent safety.
展开完整摘要收起摘要↓
LLM agents increasingly operate through multi-turn tool use and environment interaction, where safety risks often emerge from intermediate steps long before they surface in the final outcome. Reactive auditing is therefore insufficient: post-hoc diagnosis frequently misses the chance to flag risks while they are unfolding. We propose TRACES, a representation-based proactive auditor that learns prefix-level trajectory risk states from the hidden representations of an observer LLM. TRACES induces latent mechanism features from step representations and models their temporal evolution to estimate whether a partial trajectory is drifting toward unsafe behavior. To sidestep the cost and ambiguity of step-level risk annotation, TRACES is trained with weak trajectory-level supervision while still producing dense prefix-level risk estimates. Across multiple agent safety benchmarks, TRACES improves both full-trajectory safety prediction and proactive risk discrimination. Our analyses further suggest that these risk states can help train a safer agent, highlighting the broader potential of proactive auditing for long-horizon agent safety.
作者Hanyu Wang, Yifan Lan, Bochuan Cao, Lu Lin, Jinghui Chen
Agent skills provide a lightweight way to adapt LLM agents to specialized domains by storing reusable procedural knowledge in structured files. However, whether downloaded from third parties or self-generated, these skills are often unreliable, incomplete, or outdated. Existing skill-evolution methods often address these deficiencies through heuristic reflections without an explicit optimization formulation. In this paper, we propose SkillGrad, a gradient-descent-inspired framework for optimizing agent skills. SkillGrad treats the skill package as a structured parameter to optimize in a gradient descent fashion: task executions provide trajectory-level loss evidence, automatic diagnoses then provide text-based gradients that indicate the correction directions. To stabilize optimization across iterations, a momentum agent accumulates recurring diagnostic patterns into a persistent memory overlay. Finally, an LLM-based patcher executes the parameter update by applying layer-aware edits to the skill package. Evaluated on SpreadsheetBench Verified and WikiTableQuestions, SkillGrad consistently outperforms training-based skill evolution baselines across two backbone LLMs, improving over the strongest training-based baseline by $6.7$ percentage points on average. Ablations further show that momentum and contrastive diagnosis both contribute to the final skill quality.
展开完整摘要收起摘要↓
Agent skills provide a lightweight way to adapt LLM agents to specialized domains by storing reusable procedural knowledge in structured files. However, whether downloaded from third parties or self-generated, these skills are often unreliable, incomplete, or outdated. Existing skill-evolution methods often address these deficiencies through heuristic reflections without an explicit optimization formulation. In this paper, we propose SkillGrad, a gradient-descent-inspired framework for optimizing agent skills. SkillGrad treats the skill package as a structured parameter to optimize in a gradient descent fashion: task executions provide trajectory-level loss evidence, automatic diagnoses then provide text-based gradients that indicate the correction directions. To stabilize optimization across iterations, a momentum agent accumulates recurring diagnostic patterns into a persistent memory overlay. Finally, an LLM-based patcher executes the parameter update by applying layer-aware edits to the skill package. Evaluated on SpreadsheetBench Verified and WikiTableQuestions, SkillGrad consistently outperforms training-based skill evolution baselines across two backbone LLMs, improving over the strongest training-based baseline by $6.7$ percentage points on average. Ablations further show that momentum and contrastive diagnosis both contribute to the final skill quality.
Many decentralized decision problems require multiple parties to coordinate on shared decisions while keeping objectives, constraints, and data private. Large language models (LLMs) offer a promising way to lower the barrier to participation by generating local optimization agents from natural-language specifications. In coordination settings, however, executability is not enough: a generated agent may compile, solve, and pass local checks while still being semantically wrong, for example by misrepresenting costs, mis-scoping constraints, or responding incorrectly to incentives. Such errors often surface only during coordination, as systematic behavioral failures rather than infeasibility. We propose coordination-in-the-loop verification and repair for LLM-generated optimization agents. We instantiate this idea with an Alternating Direction Method of Multipliers (ADMM)-style consensus protocol and introduce OptiLoop, a pipeline that generates local optimization agents from text, verifies them through short, bounded coordination runs against a fixed reference counterparty, extracts structured behavioral and static evidence, and applies evidence-driven repair. When failures are structural rather than implementational, OptiLoop escalates from localized code fixes to corrected-formulation repair, and it can additionally reuse episodic lessons from prior instances. On 40 held-out test scenarios, OptiLoop-Full improves objective match from 66.0% to 93.0% and social match from 68.5% to 89.0% relative to a strong local-validation baseline, while reducing mean objective gap from 15.3% to 3.5% and mean social gap from 7.6% to 2.0%. These results show that, for generated optimization agents deployed inside decentralized decision loops, correctness should be validated in the loop itself rather than through isolated execution alone.
展开完整摘要收起摘要↓
Many decentralized decision problems require multiple parties to coordinate on shared decisions while keeping objectives, constraints, and data private. Large language models (LLMs) offer a promising way to lower the barrier to participation by generating local optimization agents from natural-language specifications. In coordination settings, however, executability is not enough: a generated agent may compile, solve, and pass local checks while still being semantically wrong, for example by misrepresenting costs, mis-scoping constraints, or responding incorrectly to incentives. Such errors often surface only during coordination, as systematic behavioral failures rather than infeasibility. We propose coordination-in-the-loop verification and repair for LLM-generated optimization agents. We instantiate this idea with an Alternating Direction Method of Multipliers (ADMM)-style consensus protocol and introduce OptiLoop, a pipeline that generates local optimization agents from text, verifies them through short, bounded coordination runs against a fixed reference counterparty, extracts structured behavioral and static evidence, and applies evidence-driven repair. When failures are structural rather than implementational, OptiLoop escalates from localized code fixes to corrected-formulation repair, and it can additionally reuse episodic lessons from prior instances. On 40 held-out test scenarios, OptiLoop-Full improves objective match from 66.0% to 93.0% and social match from 68.5% to 89.0% relative to a strong local-validation baseline, while reducing mean objective gap from 15.3% to 3.5% and mean social gap from 7.6% to 2.0%. These results show that, for generated optimization agents deployed inside decentralized decision loops, correctness should be validated in the loop itself rather than through isolated execution alone.
Formal verification offers a path to provably correct software, but writing verified code remains expensive enough that the technique is rarely used in production. Recent large language models can accelerate this work, and recent benchmarks measure their ability to translate specifications into code and machine-checked proofs of correctness. This thesis evaluates the state of such LLM-driven verified-code generation ("vericoding") in Lean and develops search-based methods for improving verification performance. We first reproduce a subset of the vericoding-benchmark Lean leaderboard on a current cross-vendor model pool, finding that non-reasoning performance remains roughly steady on US closed-source models while open-weight models have slightly improved. We update the iterative methodology of vericoding-benchmark with an agentic loop equipped with mathlib search, finding that model performance greatly improves and scales with agent budget. GPT-5.4 nearly saturates the benchmark at 95.0% on 423 specs with $K=50$ LLM calls. We then design two agent-directed tree-search formulations: a state-based orchestrator that branches on partial-proof states, and a context-based orchestrator that branches on full subagent contexts. Compared against the agent baseline, the context-based design solves a wider range of intermediate-difficulty specs at lower token cost, while the agent baseline retains an advantage on the hardest specs, where uninterrupted iteration matters most. We conclude that search structure has selective advantages over a strong agent baseline, and that more challenging benchmarks drawn from modern code are important to measure and drive further progress in automated formal verification. Code available upon request by contacting the author at leoy@mit.edu.
展开完整摘要收起摘要↓
Formal verification offers a path to provably correct software, but writing verified code remains expensive enough that the technique is rarely used in production. Recent large language models can accelerate this work, and recent benchmarks measure their ability to translate specifications into code and machine-checked proofs of correctness. This thesis evaluates the state of such LLM-driven verified-code generation ("vericoding") in Lean and develops search-based methods for improving verification performance. We first reproduce a subset of the vericoding-benchmark Lean leaderboard on a current cross-vendor model pool, finding that non-reasoning performance remains roughly steady on US closed-source models while open-weight models have slightly improved. We update the iterative methodology of vericoding-benchmark with an agentic loop equipped with mathlib search, finding that model performance greatly improves and scales with agent budget. GPT-5.4 nearly saturates the benchmark at 95.0% on 423 specs with $K=50$ LLM calls. We then design two agent-directed tree-search formulations: a state-based orchestrator that branches on partial-proof states, and a context-based orchestrator that branches on full subagent contexts. Compared against the agent baseline, the context-based design solves a wider range of intermediate-difficulty specs at lower token cost, while the agent baseline retains an advantage on the hardest specs, where uninterrupted iteration matters most. We conclude that search structure has selective advantages over a strong agent baseline, and that more challenging benchmarks drawn from modern code are important to measure and drive further progress in automated formal verification. Code available upon request by contacting the author at leoy@mit.edu.
作者Yuchen Guo, Junli Gong, Hongmin Cai, Yiu-ming Cheung, Weifeng Su
We present PEAM, a Parametric Embodied Agent Memory framework in Minecraft that transforms agent memory from inference-time retrieval into parameter-resident skills internalized through experience. PEAM pairs a slow deliberative LLM for open-ended reasoning with a fast parametric module for reflexive execution of consolidated skills. The fast module is a multimodal Mixture-of-Experts LoRA architecture with per-category physically isolated adapters, enabling parameter-level continual learning without catastrophic forgetting. We treat failure as a first-class training signal: failure--correction trajectory pairs are internalized through a joint behavioral-cloning and contrastive objective, so the agent learns not only what succeeds but also how corrected actions differ from failed ones. To govern consolidation, PEAM introduces a parameterization-worthiness score for deciding which experience should be internalized, and a scale-free self-triggered consolidation mechanism for deciding when to internalize without task-specific hand-tuned thresholds, making the agent self-evolving as the trigger transfers across task distributions without re-tuning. Experiments in Minecraft show that PEAM improves long-horizon task performance, mitigates forgetting on previously consolidated skills, and improves parametric-versus-retrieval efficiency over retrieval-based embodied agents and parametric memory variants.
展开完整摘要收起摘要↓
We present PEAM, a Parametric Embodied Agent Memory framework in Minecraft that transforms agent memory from inference-time retrieval into parameter-resident skills internalized through experience. PEAM pairs a slow deliberative LLM for open-ended reasoning with a fast parametric module for reflexive execution of consolidated skills. The fast module is a multimodal Mixture-of-Experts LoRA architecture with per-category physically isolated adapters, enabling parameter-level continual learning without catastrophic forgetting. We treat failure as a first-class training signal: failure--correction trajectory pairs are internalized through a joint behavioral-cloning and contrastive objective, so the agent learns not only what succeeds but also how corrected actions differ from failed ones. To govern consolidation, PEAM introduces a parameterization-worthiness score for deciding which experience should be internalized, and a scale-free self-triggered consolidation mechanism for deciding when to internalize without task-specific hand-tuned thresholds, making the agent self-evolving as the trigger transfers across task distributions without re-tuning. Experiments in Minecraft show that PEAM improves long-horizon task performance, mitigates forgetting on previously consolidated skills, and improves parametric-versus-retrieval efficiency over retrieval-based embodied agents and parametric memory variants.
作者Cihan Xiao, Yiwen Shao, Chenxing Li, Xiang He, Zhenwen Liang, Steve Yves, Sanjeev Khudanpur, Liefeng Bo
Audio and omni-modal large language models exhibit impressive cross-modal reasoning capabilities. However, applying standard reinforcement learning post-training algorithms to these models exposes a critical structural vulnerability: methods like GRPO apply uniform policy gradients across all tokens, ignoring their unequal dependence on the non-text source modality. This exacerbates late-stage modality collapse during extended chain-of-thought generation, where models progressively abandon the primary source signal in favor of compressed textual priors, leading to confident but ungrounded hallucinations. To address this, we introduce Modality-Aware Policy Optimization (MAPO), a novel dual-branch reinforcement learning framework. First, MAPO dynamically concentrates the policy gradient on modality-critical tokens using a modality relevance mask, which is derived from the cross-modal differential entropy between an audio-ablated reference and the multimodal policy. Second, it integrates an auxiliary attention loss branch that applies a targeted, temporally scaled penalty to the model's internal attention distributions. This ensures the model actively sustains cross-modal grounding deep into the reasoning trace. Evaluations on complex audio reasoning benchmarks demonstrate that MAPO substantially improves long-horizon reasoning fidelity and multimodal instruction following, achieving highly competitive performance and setting new state-of-the-art results on several key benchmarks among open-weight models. By relying strictly on native statistical signals rather than domain-specific inductive biases, MAPO offers a promising foundation for mitigating epistemic collapse across diverse multimodal systems.
展开完整摘要收起摘要↓
Audio and omni-modal large language models exhibit impressive cross-modal reasoning capabilities. However, applying standard reinforcement learning post-training algorithms to these models exposes a critical structural vulnerability: methods like GRPO apply uniform policy gradients across all tokens, ignoring their unequal dependence on the non-text source modality. This exacerbates late-stage modality collapse during extended chain-of-thought generation, where models progressively abandon the primary source signal in favor of compressed textual priors, leading to confident but ungrounded hallucinations. To address this, we introduce Modality-Aware Policy Optimization (MAPO), a novel dual-branch reinforcement learning framework. First, MAPO dynamically concentrates the policy gradient on modality-critical tokens using a modality relevance mask, which is derived from the cross-modal differential entropy between an audio-ablated reference and the multimodal policy. Second, it integrates an auxiliary attention loss branch that applies a targeted, temporally scaled penalty to the model's internal attention distributions. This ensures the model actively sustains cross-modal grounding deep into the reasoning trace. Evaluations on complex audio reasoning benchmarks demonstrate that MAPO substantially improves long-horizon reasoning fidelity and multimodal instruction following, achieving highly competitive performance and setting new state-of-the-art results on several key benchmarks among open-weight models. By relying strictly on native statistical signals rather than domain-specific inductive biases, MAPO offers a promising foundation for mitigating epistemic collapse across diverse multimodal systems.
We characterize the pre-softmax attention matrix $\mathbf{QK^\top}$ in transformers as an associative memory matrix encoding pairwise associations between input features. By decomposing this matrix into its symmetric and skew-symmetric parts, we interpret the symmetric component as governing the structure of the energy landscape, and the skew-symmetric component as driving circulation on that landscape. Leveraging the energy formulation induced by the symmetric component, we derive Hopfield-style stability measures that quantify the stability of retrieved features. We observe meaningful correlations between Hopfield-style stability measures and the fidelity-diversity trade-offs in generation. Finally, we propose a controllable knob to modulate this trade-off by modifying the circulation of the underlying dynamics. Code is available at our GitHub (https://github.com/hyeon-cho/Attention-Symmetric-Decomposition).
展开完整摘要收起摘要↓
We characterize the pre-softmax attention matrix $\mathbf{QK^\top}$ in transformers as an associative memory matrix encoding pairwise associations between input features. By decomposing this matrix into its symmetric and skew-symmetric parts, we interpret the symmetric component as governing the structure of the energy landscape, and the skew-symmetric component as driving circulation on that landscape. Leveraging the energy formulation induced by the symmetric component, we derive Hopfield-style stability measures that quantify the stability of retrieved features. We observe meaningful correlations between Hopfield-style stability measures and the fidelity-diversity trade-offs in generation. Finally, we propose a controllable knob to modulate this trade-off by modifying the circulation of the underlying dynamics. Code is available at our GitHub (https://github.com/hyeon-cho/Attention-Symmetric-Decomposition).
Safety evaluations of language models often treat serving configuration as fixed background infrastructure, but batch condition is an untested treatment variable whenever the same prompt may be evaluated alone, in a synchronized batch, or inside a continuous-batching scheduler. We synthesize four artifact-backed studies into a paired testing protocol: Study A combines local discovery, scorer-corrected adjudication, and true-batching confirmation; Study B tests cross-model generalization; Study C tests continuous-batch composition; and Study D runs a batch-invariant-kernel ablation. The local test finds safety-label changes more often than capability-label changes (0.51% vs. 0.14%), but adjudication of 63 candidate rows leaves only 17 genuine behavioral flips, implying a corrected full-set rate of 0.16%. The 15-model extension finds no detectable universal safety-over-capability skew: flips are near parity (0.94x), alignment type has no detectable association ($p=0.942$, $η^2=0.033$), and output instability is the strongest tested fragility screen ($r=0.909$, bootstrap 95% CI [0.65, 0.97]). In the targeted kernel ablation, standard vLLM reproduces 22/55 label flips on current score-flip candidates, while enabling VLLM_BATCH_INVARIANT=1 reduces the same test to 0/55 flips; the composition test separately finds no aggregate effect at 4.7pp sensitivity. The testing recommendation is exact-stack validation: evaluate refusal at the served batch setting, pair safety prompts with capability controls, and report low-rate directional flips separately from aggregate null effects.
展开完整摘要收起摘要↓
Safety evaluations of language models often treat serving configuration as fixed background infrastructure, but batch condition is an untested treatment variable whenever the same prompt may be evaluated alone, in a synchronized batch, or inside a continuous-batching scheduler. We synthesize four artifact-backed studies into a paired testing protocol: Study A combines local discovery, scorer-corrected adjudication, and true-batching confirmation; Study B tests cross-model generalization; Study C tests continuous-batch composition; and Study D runs a batch-invariant-kernel ablation. The local test finds safety-label changes more often than capability-label changes (0.51% vs. 0.14%), but adjudication of 63 candidate rows leaves only 17 genuine behavioral flips, implying a corrected full-set rate of 0.16%. The 15-model extension finds no detectable universal safety-over-capability skew: flips are near parity (0.94x), alignment type has no detectable association ($p=0.942$, $η^2=0.033$), and output instability is the strongest tested fragility screen ($r=0.909$, bootstrap 95% CI [0.65, 0.97]). In the targeted kernel ablation, standard vLLM reproduces 22/55 label flips on current score-flip candidates, while enabling VLLM_BATCH_INVARIANT=1 reduces the same test to 0/55 flips; the composition test separately finds no aggregate effect at 4.7pp sensitivity. The testing recommendation is exact-stack validation: evaluate refusal at the served batch setting, pair safety prompts with capability controls, and report low-rate directional flips separately from aggregate null effects.
作者Jaewoo Lee, Hyeongyu Kang, Dohyun Kim, Kyuil Sim, Woocheol Shin, Minsu Kim, Taeyoung Yun, Jeongjae Lee, Sanghyeok Choi, Tabitha Edith Lee, Jong Chul Ye, Jinkyoo Park
Aligning a few-step generative model is challenging, since existing alignment frameworks typically rely on restrictive assumptions: a tractable likelihood, a specific ODE/SDE solver, or a particular model family. We introduce FAV, Few-step Generative Models Alignment via Sample-based Variational Inference, a general alignment framework that requires only sample access to the generator and the reference distribution. We cast alignment as sampling from a reward-tilted distribution anchored to a reference distribution. We leverage Stein Variational Gradient Descent as a sample-based variational inference scheme and amortize its particle updates into the generator parameters via fixed-point regression. We evaluate FAV on two domains: robotics manipulation and image generator alignment. On generative policy alignment for robotic manipulation, FAV outperforms prevailing policy extraction baselines across 56 offline and 30 offline-to-online RL tasks. For image generator alignment, FAV fine-tunes diverse few-step backbones, including GAN, drifting model, consistency models, and flow maps, scaling from ImageNet-$256$ to 1024$^2$ text-to-image synthesis. Code is available at https://github.com/Jaewoopudding/FAV.
展开完整摘要收起摘要↓
Aligning a few-step generative model is challenging, since existing alignment frameworks typically rely on restrictive assumptions: a tractable likelihood, a specific ODE/SDE solver, or a particular model family. We introduce FAV, Few-step Generative Models Alignment via Sample-based Variational Inference, a general alignment framework that requires only sample access to the generator and the reference distribution. We cast alignment as sampling from a reward-tilted distribution anchored to a reference distribution. We leverage Stein Variational Gradient Descent as a sample-based variational inference scheme and amortize its particle updates into the generator parameters via fixed-point regression. We evaluate FAV on two domains: robotics manipulation and image generator alignment. On generative policy alignment for robotic manipulation, FAV outperforms prevailing policy extraction baselines across 56 offline and 30 offline-to-online RL tasks. For image generator alignment, FAV fine-tunes diverse few-step backbones, including GAN, drifting model, consistency models, and flow maps, scaling from ImageNet-$256$ to 1024$^2$ text-to-image synthesis. Code is available at https://github.com/Jaewoopudding/FAV.
Large language model (LLM) serving creates environmental impacts beyond carbon and water, including ecosystem damage through biodiversity-related pathways. We present BIRDS, a framework for Biodiversity Impact of Request-Driven LLM Serving. BIRDS defines request-level functional units, quantifies operational and embodied biodiversity impact, and introduces Quality-Normalized Biodiversity Impact (QNBI) to jointly analyze ecological impact and response quality. Across diverse workloads, models, GPUs, and regions, BIRDS reveals that biodiversity impact accumulates at scale and exposes actionable quality-aware serving tradeoffs.
展开完整摘要收起摘要↓
Large language model (LLM) serving creates environmental impacts beyond carbon and water, including ecosystem damage through biodiversity-related pathways. We present BIRDS, a framework for Biodiversity Impact of Request-Driven LLM Serving. BIRDS defines request-level functional units, quantifies operational and embodied biodiversity impact, and introduces Quality-Normalized Biodiversity Impact (QNBI) to jointly analyze ecological impact and response quality. Across diverse workloads, models, GPUs, and regions, BIRDS reveals that biodiversity impact accumulates at scale and exposes actionable quality-aware serving tradeoffs.
Multi-agent LLM systems have become the dominant production workload, but the serving stack was not built for them. The agent framework above knows agent identities, role, schemas, and dispatch structure but never sees an engine-level event; the serving engine below sees every event but knows nothing about agents. A surprising number of cross-cutting policies depend on both: prefix caching, batch shaping, speculative execution, fairness, tool-result memoization, safety enforcement, and more. Each lives in the seam between the two layers and is currently solved by a one-off patch into one neighbor or the other. We argue this seam is best addressed by an architectural change rather than point fixes: insert a third tier, an agent runtime layer, between the framework and the engine, exposing four primitives (observe, score, predict, act) into which any agent-aware policy plugs, with agent identity as the shared coordinate. We map nine concrete policies onto the layer and validate the abstraction in depth on the one with the largest immediate serving-cost lever: KV caching across sessions, instantiated as CacheSage, which learns the per-workload agent transition matrix online and uses it for survival-based eviction and between-step prefetch. Preliminary results on five real multi-agent workloads show +13 to +37 pp cache hit-rate lift, 12% to 29% lower mean TTFT, and 6% to 14% higher throughput over an unmodified serving stack.
展开完整摘要收起摘要↓
Multi-agent LLM systems have become the dominant production workload, but the serving stack was not built for them. The agent framework above knows agent identities, role, schemas, and dispatch structure but never sees an engine-level event; the serving engine below sees every event but knows nothing about agents. A surprising number of cross-cutting policies depend on both: prefix caching, batch shaping, speculative execution, fairness, tool-result memoization, safety enforcement, and more. Each lives in the seam between the two layers and is currently solved by a one-off patch into one neighbor or the other. We argue this seam is best addressed by an architectural change rather than point fixes: insert a third tier, an agent runtime layer, between the framework and the engine, exposing four primitives (observe, score, predict, act) into which any agent-aware policy plugs, with agent identity as the shared coordinate. We map nine concrete policies onto the layer and validate the abstraction in depth on the one with the largest immediate serving-cost lever: KV caching across sessions, instantiated as CacheSage, which learns the per-workload agent transition matrix online and uses it for survival-based eviction and between-step prefetch. Preliminary results on five real multi-agent workloads show +13 to +37 pp cache hit-rate lift, 12% to 29% lower mean TTFT, and 6% to 14% higher throughput over an unmodified serving stack.
Modern retrieval-augmented generation(RAG) deployments increasingly rely on caching to reduce token cost and time-to-first-token(TTFT). Prefix-level KV reuse is now standard in serving stacks such as vLLM, and chunk-level and position-independent reuse have been pushed further by recent systems(RAGCache, TurboRAG, CacheBlend, EPIC, ContextPilot, PCR, LMCache). Output-level semantic answer caches, by contrast, remain fragile: similar prompts can map to different correct answers, retrieved evidence drifts as the corpus is updated, and adversarial collision attacks have been shown to hijack cached responses. We argue that the right framing for cached answer reuse is not how to reuse faster but when reuse is safe. We propose GroundedCache, an evidence-validated cache router that admits a cached answer only when 4 cheap gates simultaneously hold: query similarity, retrieved-evidence overlap, source-version validity, and lexical (or judge-based) support of the cached answer by the freshly retrieved evidence. We build a six-regime workload that stress-tests cache safety rather than only hit rate, and introduce an operator-facing metric, the unsafe-served rate (USR), fraction of all queries that received a wrong cached answer. Across 2 datasets and 12,000 real-LLM generations(Qwen2.5-7B-Instruct on vLLM with Automatic Prefix Caching), GroundedCache drives USR to 0.0% on every HotpotQA regime(vs. 15-35% under naive caching) and to 1.5% on mtRAG document drift(vs. 51.5%), a 34x reduction on the design-point adversarial regime and 3-10x reductions across the other mtRAG regimes, while end-to-end p50 latency stays within 1.04-1.07x of a no-cache RAG baseline. A per-gate ablation isolates the lexical support gate as the load-bearing safety mechanism on both datasets, with the remaining gates providing defense-in-depth at near-zero cost. We release the implementation, workload, and evaluation harness.
展开完整摘要收起摘要↓
Modern retrieval-augmented generation(RAG) deployments increasingly rely on caching to reduce token cost and time-to-first-token(TTFT). Prefix-level KV reuse is now standard in serving stacks such as vLLM, and chunk-level and position-independent reuse have been pushed further by recent systems(RAGCache, TurboRAG, CacheBlend, EPIC, ContextPilot, PCR, LMCache). Output-level semantic answer caches, by contrast, remain fragile: similar prompts can map to different correct answers, retrieved evidence drifts as the corpus is updated, and adversarial collision attacks have been shown to hijack cached responses. We argue that the right framing for cached answer reuse is not how to reuse faster but when reuse is safe. We propose GroundedCache, an evidence-validated cache router that admits a cached answer only when 4 cheap gates simultaneously hold: query similarity, retrieved-evidence overlap, source-version validity, and lexical (or judge-based) support of the cached answer by the freshly retrieved evidence. We build a six-regime workload that stress-tests cache safety rather than only hit rate, and introduce an operator-facing metric, the unsafe-served rate (USR), fraction of all queries that received a wrong cached answer. Across 2 datasets and 12,000 real-LLM generations(Qwen2.5-7B-Instruct on vLLM with Automatic Prefix Caching), GroundedCache drives USR to 0.0% on every HotpotQA regime(vs. 15-35% under naive caching) and to 1.5% on mtRAG document drift(vs. 51.5%), a 34x reduction on the design-point adversarial regime and 3-10x reductions across the other mtRAG regimes, while end-to-end p50 latency stays within 1.04-1.07x of a no-cache RAG baseline. A per-gate ablation isolates the lexical support gate as the load-bearing safety mechanism on both datasets, with the remaining gates providing defense-in-depth at near-zero cost. We release the implementation, workload, and evaluation harness.
作者Kabir Swain, Sijie Han, Daniel Karl I. Weidele, Mauro Martino, David Cox, Antonio Torralba
We propose Hurwitz Quaternion Multiplicative Quantization (HQMQ), a calibration-free method for KV cache compression of large language models. HQMQ treats each 4-element chunk of K or V as a quaternion and quantizes its unit direction to the product $q_p \cdot q_s$, where $q_p$ ranges over the 24-element Hurwitz group $2T$ (the 24 vertices of the 24-cell on $S^3$, pairwise angle $60^\circ$) and $q_s$ ranges over a per-(layer, head) secondary codebook of $S$ random unit quaternions. The multiplicative composition yields $24S$ effective codewords at $S$ stored parameters; random initialization suffices because left-multiplication is an $S^3$ isometry, so seeded codebooks vary in end-task ppl by $<1.5%$. A per-batch median-multiplier outlier extraction step ($C{=}3$, no calibration) handles modern outlier-heavy architectures. We evaluate on five modern open models: Mistral-7B (dense MHA), Llama-3-8B and Qwen2.5-7B and Qwen3-8B (dense GQA), and gpt-oss-20b (sparse MoE). On Mistral-7B and Qwen3-8B, HQMQ matches fp16 within $0.02$--$0.03$ ppl points at $\sim$5 bits. On Qwen2.5-7B and Qwen3-8B, where naive int4 collapses to $10^4{+}$ ppl, HQMQ + Med3$\times$ recovers fp16 quality within $0.02$--$0.10$ ppl points at $\sim$5 bits. HQMQ Pareto-dominates naive int by $3$--$1900\times$ at matched bits across all five models, and downstream zero-shot accuracy matches fp16 at $3.79$ bits on Mistral. Against the strongest calibrated KV-quantization baseline, HQMQ at $3.79$ bits matches KIVI-4 ($\sim 4.5$ bits) within ${\sim}1$ pt on CoQA, $0.6$ pts on TruthfulQA, and $2.3$ pts on GSM8K, at $16%$ fewer bits and without a calibration pass. At the storage level, HQMQ delivers up to $5.05\times$ KV compression, shrinking a Llama-3-70B 128k-context cache from 43 GB to 8.5 GB.
展开完整摘要收起摘要↓
We propose Hurwitz Quaternion Multiplicative Quantization (HQMQ), a calibration-free method for KV cache compression of large language models. HQMQ treats each 4-element chunk of K or V as a quaternion and quantizes its unit direction to the product $q_p \cdot q_s$, where $q_p$ ranges over the 24-element Hurwitz group $2T$ (the 24 vertices of the 24-cell on $S^3$, pairwise angle $60^\circ$) and $q_s$ ranges over a per-(layer, head) secondary codebook of $S$ random unit quaternions. The multiplicative composition yields $24S$ effective codewords at $S$ stored parameters; random initialization suffices because left-multiplication is an $S^3$ isometry, so seeded codebooks vary in end-task ppl by $<1.5%$. A per-batch median-multiplier outlier extraction step ($C{=}3$, no calibration) handles modern outlier-heavy architectures. We evaluate on five modern open models: Mistral-7B (dense MHA), Llama-3-8B and Qwen2.5-7B and Qwen3-8B (dense GQA), and gpt-oss-20b (sparse MoE). On Mistral-7B and Qwen3-8B, HQMQ matches fp16 within $0.02$--$0.03$ ppl points at $\sim$5 bits. On Qwen2.5-7B and Qwen3-8B, where naive int4 collapses to $10^4{+}$ ppl, HQMQ + Med3$\times$ recovers fp16 quality within $0.02$--$0.10$ ppl points at $\sim$5 bits. HQMQ Pareto-dominates naive int by $3$--$1900\times$ at matched bits across all five models, and downstream zero-shot accuracy matches fp16 at $3.79$ bits on Mistral. Against the strongest calibrated KV-quantization baseline, HQMQ at $3.79$ bits matches KIVI-4 ($\sim 4.5$ bits) within ${\sim}1$ pt on CoQA, $0.6$ pts on TruthfulQA, and $2.3$ pts on GSM8K, at $16%$ fewer bits and without a calibration pass. At the storage level, HQMQ delivers up to $5.05\times$ KV compression, shrinking a Llama-3-70B 128k-context cache from 43 GB to 8.5 GB.
We describe our preliminary design of a real-time asynchronous event-based monocular odometry for planetary exploration. Operating under strict computational constraints, planetary rovers frequently encounter complex, unpredictable environments that demand high-speed sensing and robustness to high dynamic range (HDR) lighting. Event cameras address these needs by reporting asynchronous, pixel-wise brightness changes with microsecond resolution, significantly reducing data bandwidth while maintaining robustness in extreme lighting conditions. We propose an approach based on an Error-State Kalman Filter (ESKF) that leverages this asynchronous event stream to continuously estimate camera ego-motion. The camera state is updated with every tracked position output generated by RATE, a real-time asynchronous feature tracker.
展开完整摘要收起摘要↓
We describe our preliminary design of a real-time asynchronous event-based monocular odometry for planetary exploration. Operating under strict computational constraints, planetary rovers frequently encounter complex, unpredictable environments that demand high-speed sensing and robustness to high dynamic range (HDR) lighting. Event cameras address these needs by reporting asynchronous, pixel-wise brightness changes with microsecond resolution, significantly reducing data bandwidth while maintaining robustness in extreme lighting conditions. We propose an approach based on an Error-State Kalman Filter (ESKF) that leverages this asynchronous event stream to continuously estimate camera ego-motion. The camera state is updated with every tracked position output generated by RATE, a real-time asynchronous feature tracker.
Modern analytics systems are fundamentally reactive, requiring users to define queries over increasingly complex and continuously evolving data. In real-time streaming environments, this paradigm breaks down, as the space of potential insights becomes too large to enumerate manually. We present a multi-agent architecture for autonomous insight discovery over real-time data streams. The system implements a continuous discovery loop in which agents generate hypotheses, compile them into executable analytics, validate generated artifacts, and produce visualizations and deployable applications. The architecture leverages Apache Kafka for event-driven coordination, Apache Flink for stream processing, and large language models to implement specialized agents. A key contribution is a contract-driven design based on typed intermediate artifacts, enabling modularity, observability, lineage, and safer execution of dynamically generated analytics. Through use cases in retail, finance, and public data, we show how this architecture supports a shift from query-driven analytics to proactive, discovery-driven systems.
展开完整摘要收起摘要↓
Modern analytics systems are fundamentally reactive, requiring users to define queries over increasingly complex and continuously evolving data. In real-time streaming environments, this paradigm breaks down, as the space of potential insights becomes too large to enumerate manually. We present a multi-agent architecture for autonomous insight discovery over real-time data streams. The system implements a continuous discovery loop in which agents generate hypotheses, compile them into executable analytics, validate generated artifacts, and produce visualizations and deployable applications. The architecture leverages Apache Kafka for event-driven coordination, Apache Flink for stream processing, and large language models to implement specialized agents. A key contribution is a contract-driven design based on typed intermediate artifacts, enabling modularity, observability, lineage, and safer execution of dynamically generated analytics. Through use cases in retail, finance, and public data, we show how this architecture supports a shift from query-driven analytics to proactive, discovery-driven systems.
作者Boxiang Qiu, Liliang Chen, Yue Liao, Nan Wang, Lintao Wang, Jiayi Luo, Wenzhi Zhao, Shengcong Chen, Di Chen, Ye Li, Chen Gao, Shuicheng Yan, Si Liu, Maoqing Yao, Guanghui Ren
We introduce GE-Sim 2.0 (Genie Envisioner World Simulator 2.0), a closed-loop video world simulator for robotic manipulation. Building on the action-conditioned video generation framework of Genie Envisioner, GE-Sim 2.0 is re-trained on thousands of hours of real-world robot data spanning teleoperation, contact-rich interaction, and on-robot policy deployment, substantially improving action-following fidelity and trajectory coverage. On top of this foundation, three new modules close the loop from video simulation to policy learning: a state expert that decodes proprioceptive state from video latents to support next-chunk prediction by downstream VLA policies; a world judge that scores generated rollouts against task instructions, yielding machine-verifiable success signals and rewards in place of manual inspection; and an acceleration framework that delivers a 25-frame rollout in 2.3 seconds on a single H100, with up to 4* frame skipping at inference for long-horizon evaluation. GE-Sim 2.0 tops the public WorldArena leaderboard at only 2B parameters, outperforming both dedicated robotic world models and closed-source general video generators, and policies trained against its rollouts and rewards translate into measurable real-world gains, establishing GE-Sim 2.0 as a practical platform for scalable evaluation and closed-loop learning of manipulation policies.
展开完整摘要收起摘要↓
We introduce GE-Sim 2.0 (Genie Envisioner World Simulator 2.0), a closed-loop video world simulator for robotic manipulation. Building on the action-conditioned video generation framework of Genie Envisioner, GE-Sim 2.0 is re-trained on thousands of hours of real-world robot data spanning teleoperation, contact-rich interaction, and on-robot policy deployment, substantially improving action-following fidelity and trajectory coverage. On top of this foundation, three new modules close the loop from video simulation to policy learning: a state expert that decodes proprioceptive state from video latents to support next-chunk prediction by downstream VLA policies; a world judge that scores generated rollouts against task instructions, yielding machine-verifiable success signals and rewards in place of manual inspection; and an acceleration framework that delivers a 25-frame rollout in 2.3 seconds on a single H100, with up to 4* frame skipping at inference for long-horizon evaluation. GE-Sim 2.0 tops the public WorldArena leaderboard at only 2B parameters, outperforming both dedicated robotic world models and closed-source general video generators, and policies trained against its rollouts and rewards translate into measurable real-world gains, establishing GE-Sim 2.0 as a practical platform for scalable evaluation and closed-loop learning of manipulation policies.
Video generation models are increasingly used as world simulators for tasks like driving and robotic manipulation. What matters in these settings is not whether a single video looks right, but whether the model's output changes when its input changes. We test this by giving a model two prompts describing the same scene with one physical detail varied, and checking whether the two videos diverge the way physics predicts. The wording difference between the prompts is small by design, since only one variable is changed, but the correct physical difference is not. A model that misses this can still produce two videos that each look plausible individually, and existing benchmarks score videos one at a time and cannot detect this failure. We introduce What-If World, 319 such prompt pairs built on real frames from nuScenes and DROID, organized by a taxonomy of six physical variables shared across driving and manipulation. Each pair is scored with APEO, a four-part rubric checking whether each video follows its prompt (Adherence), is physically consistent (Physics), preserves the shared scene (Environment), and ends in the correct difference (Outcome). Across nine state-of-the-art models, no system exceeds 52% on the paired score, and open-source models cluster near 28%. Every model tested fails on a large fraction of causal interventions, indicating substantial room before these models can reliably support action-conditioned simulation or model-based planning. Where models do score well, performance appears to track the visual prominence of the intervention rather than the tractability of its underlying physics. Some visually subtle interventions score as low as 14.2%, while visually pronounced ones reach 40.4%.
展开完整摘要收起摘要↓
Video generation models are increasingly used as world simulators for tasks like driving and robotic manipulation. What matters in these settings is not whether a single video looks right, but whether the model's output changes when its input changes. We test this by giving a model two prompts describing the same scene with one physical detail varied, and checking whether the two videos diverge the way physics predicts. The wording difference between the prompts is small by design, since only one variable is changed, but the correct physical difference is not. A model that misses this can still produce two videos that each look plausible individually, and existing benchmarks score videos one at a time and cannot detect this failure. We introduce What-If World, 319 such prompt pairs built on real frames from nuScenes and DROID, organized by a taxonomy of six physical variables shared across driving and manipulation. Each pair is scored with APEO, a four-part rubric checking whether each video follows its prompt (Adherence), is physically consistent (Physics), preserves the shared scene (Environment), and ends in the correct difference (Outcome). Across nine state-of-the-art models, no system exceeds 52% on the paired score, and open-source models cluster near 28%. Every model tested fails on a large fraction of causal interventions, indicating substantial room before these models can reliably support action-conditioned simulation or model-based planning. Where models do score well, performance appears to track the visual prominence of the intervention rather than the tractability of its underlying physics. Some visually subtle interventions score as low as 14.2%, while visually pronounced ones reach 40.4%.
作者Hui Dai, Ryan Teehan, Parsa Torabian, Mengye Ren
Probabilistic forecasting estimates the likelihood of uncertain future events. To improve LLM forecasting, existing methods typically learn from binary outcomes to output verbalized forecasts. However, while aggregated human forecasts contain rich information in both the crowd probability estimate and the degree of agreement among forecasters, how to utilize these signals remains underexplored. To address this, we propose the Beta-Bernoulli Calibrator (BBC), which converts an initial point estimate forecast from any model into a distribution over event likelihood, using supervision from both binary outcomes and human forecasts. BBC models event likelihood $p \sim \text{Beta}(α, β)$ and outcome $y \sim \text{Bernoulli}(p)$, with the mean as the calibrated point forecast and the variance as the epistemic uncertainty. Our results show that BBC generally provides better calibrated and more accurate forecasts than both traditional post-hoc calibration methods and models fine-tuned specifically for forecasting, while remaining lightweight and having good generalization. We also show that the epistemic uncertainty captured by BBC is a more reliable predictor of forecasting error than verbalized confidence.
展开完整摘要收起摘要↓
Probabilistic forecasting estimates the likelihood of uncertain future events. To improve LLM forecasting, existing methods typically learn from binary outcomes to output verbalized forecasts. However, while aggregated human forecasts contain rich information in both the crowd probability estimate and the degree of agreement among forecasters, how to utilize these signals remains underexplored. To address this, we propose the Beta-Bernoulli Calibrator (BBC), which converts an initial point estimate forecast from any model into a distribution over event likelihood, using supervision from both binary outcomes and human forecasts. BBC models event likelihood $p \sim \text{Beta}(α, β)$ and outcome $y \sim \text{Bernoulli}(p)$, with the mean as the calibrated point forecast and the variance as the epistemic uncertainty. Our results show that BBC generally provides better calibrated and more accurate forecasts than both traditional post-hoc calibration methods and models fine-tuned specifically for forecasting, while remaining lightweight and having good generalization. We also show that the epistemic uncertainty captured by BBC is a more reliable predictor of forecasting error than verbalized confidence.
作者Joan Vendrell Gallart, Russell Bent, Michael Grosskopf
Large Language Models are increasingly deployed inside agentic systems, where they must follow structured protocols, adapt to evolving states, and operate under memory, latency, and cost constraints. In such regimes, prompt extension is unreliable: growing contexts can push compact models outside their effective prompt domain, while deployment-time fine-tuning remains limited by scarce data and compute. We propose a hierarchical control-and-learning framework in which a compact model is first distilled to learn the required output schema, then supervised online by an oracle-controller loop. The controller monitors protocol validity and semantic performance, projects accumulated histories into a feasible prompt domain, and triggers lightweight oracle-supervised fine-tuning under drift. This separates schema learning for communication compatibility from semantic adaptation for task-level correction. We formalize prompt-domain feasibility and attention-induced saturation, motivating control of the effective prompt state rather than reliance on nominal context length. Using Multi-Fidelity Bayesian Optimization as a controlled sequential testbed, we characterize a core deployment failure mode and show improved reliability and cost-efficiency over non-hierarchical, distillation-only, and non-distilled baselines.
展开完整摘要收起摘要↓
Large Language Models are increasingly deployed inside agentic systems, where they must follow structured protocols, adapt to evolving states, and operate under memory, latency, and cost constraints. In such regimes, prompt extension is unreliable: growing contexts can push compact models outside their effective prompt domain, while deployment-time fine-tuning remains limited by scarce data and compute. We propose a hierarchical control-and-learning framework in which a compact model is first distilled to learn the required output schema, then supervised online by an oracle-controller loop. The controller monitors protocol validity and semantic performance, projects accumulated histories into a feasible prompt domain, and triggers lightweight oracle-supervised fine-tuning under drift. This separates schema learning for communication compatibility from semantic adaptation for task-level correction. We formalize prompt-domain feasibility and attention-induced saturation, motivating control of the effective prompt state rather than reliance on nominal context length. Using Multi-Fidelity Bayesian Optimization as a controlled sequential testbed, we characterize a core deployment failure mode and show improved reliability and cost-efficiency over non-hierarchical, distillation-only, and non-distilled baselines.
作者Yashaswi Karnati, Kamran Jafari, Akash Mehra, Li Ding, Pranav Prashant Thombre, Ali Roshan Ghias, Shifang Xu, Parth Mannan, Yu Yao, Hao Wu, Eric Harper, Ashwath Aithal, Nima Tajbakhsh
Foundation model training is becoming multimodal, from post-training pipelines to large-scale pretraining. As modality coverage broadens, context windows grow, and encoder LLM scales diverge, a single LLM-centric TP/CP/PP/DP/EP layout increasingly limits throughput. This coupling forces encoders to inherit LLM-driven sharding and placement choices that can add communication, limit encoder parallelism, or constrain the LLM schedule; the mismatch is most pronounced at long contexts, where LLM context parallelism is needed for the fused multimodal sequence but encoder inputs remain bounded. We present heterogeneous parallelism for multimodal large language model training, an abstraction that lets modules in one end-to-end graph use independent layouts and rank placements, supporting colocated execution on shared GPUs and non-colocated execution on disjoint rank sets. The key challenge is preserving boundary tensor semantics across independent layouts: forward activations must be materialized for the destination layout, while backward gradients must be routed back to the source layout. We address this with boundary communicators that implement forward and backward layout transforms, plus scheduling extensions for both placement modes. We evaluate optimized homogeneous, colocated heterogeneous, and non-colocated heterogeneous configurations across multimodal workloads and GPU scales to characterize when added layout and placement freedom exposes a better operating point. Across this sweep, colocated heterogeneity improves TFLOPS/GPU by up to 49.3%, while non-colocated heterogeneity improves aggregate token throughput by up to 13.0% and TFLOPS/GPU by up to 9.6%. We validate loss convergence parity against homogeneous baselines and release the system as an open-source Megatron-LM extension.
展开完整摘要收起摘要↓
Foundation model training is becoming multimodal, from post-training pipelines to large-scale pretraining. As modality coverage broadens, context windows grow, and encoder LLM scales diverge, a single LLM-centric TP/CP/PP/DP/EP layout increasingly limits throughput. This coupling forces encoders to inherit LLM-driven sharding and placement choices that can add communication, limit encoder parallelism, or constrain the LLM schedule; the mismatch is most pronounced at long contexts, where LLM context parallelism is needed for the fused multimodal sequence but encoder inputs remain bounded. We present heterogeneous parallelism for multimodal large language model training, an abstraction that lets modules in one end-to-end graph use independent layouts and rank placements, supporting colocated execution on shared GPUs and non-colocated execution on disjoint rank sets. The key challenge is preserving boundary tensor semantics across independent layouts: forward activations must be materialized for the destination layout, while backward gradients must be routed back to the source layout. We address this with boundary communicators that implement forward and backward layout transforms, plus scheduling extensions for both placement modes. We evaluate optimized homogeneous, colocated heterogeneous, and non-colocated heterogeneous configurations across multimodal workloads and GPU scales to characterize when added layout and placement freedom exposes a better operating point. Across this sweep, colocated heterogeneity improves TFLOPS/GPU by up to 49.3%, while non-colocated heterogeneity improves aggregate token throughput by up to 13.0% and TFLOPS/GPU by up to 9.6%. We validate loss convergence parity against homogeneous baselines and release the system as an open-source Megatron-LM extension.
作者Tamerlan Aghayev, Maxime Elkael, Michele Polese, Minh Dat Nguyen, Gabriele Gemmi, Andrea Lacava, Ali Saeizadeh, Reshma Prasad, Paolo Testolina, Angelo Feraudo, Soumendra Nanda, Pedram Johari, Salvatore D'Oro, Tommaso Melodia
Cellular research and development (R&D) is throttled by six structural processes that each consume months of manual engineering work per iteration: (i) synthesizing new features from standards or research papers into production code; (ii) conformance and interoperability testing; (iii) hardening against field anomalies and diverse deployment environments; (iv) data-driven optimization of network functionalities; (v) discovering and prototyping novel waveforms, functionalities, and capabilities for future standards; and (vi) securing the stack against vulnerabilities. Although Large Language Models (LLMs) have compressed comparable R&D work in general software engineering from days to minutes, their known pitfalls worsen on Radio Access Network (RAN) use cases: they hallucinate Application Programming Interfaces (APIs) and mis-read specifications, which kills interoperability of RAN components at the first mistake, and they heavily rely on simulations for designing algorithms, which is notorious for breaking when transferred to real hardware. To address these challenges, we present GENESIS, an agentic Artificial Intelligence (AI) framework that converts intents (e.g., a specification clause, a telemetry anomaly, or a research hypothesis) into solutions validated with over-the-air experiments, fed back into a persistent knowledge base. GENESIS is built on three composable primitives (agents, skills, hooks) and a knowledge layer (SYNAPSE) that doubles as the source of ground truth and the recipient of every artifact the framework produces, making capabilities compound across runs.
展开完整摘要收起摘要↓
Cellular research and development (R&D) is throttled by six structural processes that each consume months of manual engineering work per iteration: (i) synthesizing new features from standards or research papers into production code; (ii) conformance and interoperability testing; (iii) hardening against field anomalies and diverse deployment environments; (iv) data-driven optimization of network functionalities; (v) discovering and prototyping novel waveforms, functionalities, and capabilities for future standards; and (vi) securing the stack against vulnerabilities. Although Large Language Models (LLMs) have compressed comparable R&D work in general software engineering from days to minutes, their known pitfalls worsen on Radio Access Network (RAN) use cases: they hallucinate Application Programming Interfaces (APIs) and mis-read specifications, which kills interoperability of RAN components at the first mistake, and they heavily rely on simulations for designing algorithms, which is notorious for breaking when transferred to real hardware. To address these challenges, we present GENESIS, an agentic Artificial Intelligence (AI) framework that converts intents (e.g., a specification clause, a telemetry anomaly, or a research hypothesis) into solutions validated with over-the-air experiments, fed back into a persistent knowledge base. GENESIS is built on three composable primitives (agents, skills, hooks) and a knowledge layer (SYNAPSE) that doubles as the source of ground truth and the recipient of every artifact the framework produces, making capabilities compound across runs.
The token-level extractive compressors widely used for general LM context are structurally inappropriate for LLM agents: across 17 (env, backbone, method) cells spanning two independent token-level method families, every cell collapses to mean reward <= 0.05 despite 1.3-13.3x realized compression. We name and characterize this failure mode as action-grammar destruction — the tokens carrying action semantics (identifiers, brackets, action verbs) are exactly those self-information ranks lowest, so a general-purpose compressor reliably removes them and the environment rejects the residual. The diagnosis points to step-granularity compression. We introduce AGORA, an inference-free step-level compressor combining a structural prompt parser, an always-keep floor for format- and recency-critical content, and a 125M-parameter relevance scorer trained on counterfactual next-action-change labels (~2ms/step, zero per-step LLM toll). Across the compared inference-free and LLM-based methods, AGORA is the only one retaining >= 75% uncompressed performance in 8 of 9 cells (with the lone exception at 73%); a four-way component ablation isolates the structural floor as the dominant quality lever and the learned scorer as the source of 1.0-11.5x adaptive end-to-end compression from a single fixed keep ratio.
展开完整摘要收起摘要↓
The token-level extractive compressors widely used for general LM context are structurally inappropriate for LLM agents: across 17 (env, backbone, method) cells spanning two independent token-level method families, every cell collapses to mean reward <= 0.05 despite 1.3-13.3x realized compression. We name and characterize this failure mode as action-grammar destruction — the tokens carrying action semantics (identifiers, brackets, action verbs) are exactly those self-information ranks lowest, so a general-purpose compressor reliably removes them and the environment rejects the residual. The diagnosis points to step-granularity compression. We introduce AGORA, an inference-free step-level compressor combining a structural prompt parser, an always-keep floor for format- and recency-critical content, and a 125M-parameter relevance scorer trained on counterfactual next-action-change labels (~2ms/step, zero per-step LLM toll). Across the compared inference-free and LLM-based methods, AGORA is the only one retaining >= 75% uncompressed performance in 8 of 9 cells (with the lone exception at 73%); a four-way component ablation isolates the structural floor as the dominant quality lever and the learned scorer as the source of 1.0-11.5x adaptive end-to-end compression from a single fixed keep ratio.
Modern retrieval agents expose many configuration choices — LLM, retriever, number of documents, number of hops, and synthesis strategy — each shaping both answer quality and serving cost. Today, these pipelines are typically hand-tuned once per workload, leaving substantial per-query optimization untapped. We formulate the problem: given a natural-language query and either an accuracy or a budget target, select from a predefined pipeline catalog the configuration that minimizes cost or maximizes accuracy at inference time. We propose **BRANE**, which uses an LLM to convert each query into workload-specific characteristics, then trains a lightweight per-configuration predictor that estimates whether the pipeline will answer the query correctly. At inference time, **BRANE** selects the configuration that maximizes predicted correctness penalized by cost, exposing a tunable cost-quality tradeoff without retraining. Across MuSiQue, BrowseComp-Plus, and FinanceBench, **BRANE** consistently pushes the cost-quality Pareto frontier, matches the best fixed configuration's accuracy at up to 89% lower cost, and outperforms LLM-routing, rule-based, and fine-tuned Qwen3-4B baselines. These results show that per-query configuration of the full retrieval pipeline is a practical alternative to static workload-level tuning.
展开完整摘要收起摘要↓
Modern retrieval agents expose many configuration choices — LLM, retriever, number of documents, number of hops, and synthesis strategy — each shaping both answer quality and serving cost. Today, these pipelines are typically hand-tuned once per workload, leaving substantial per-query optimization untapped. We formulate the problem: given a natural-language query and either an accuracy or a budget target, select from a predefined pipeline catalog the configuration that minimizes cost or maximizes accuracy at inference time. We propose **BRANE**, which uses an LLM to convert each query into workload-specific characteristics, then trains a lightweight per-configuration predictor that estimates whether the pipeline will answer the query correctly. At inference time, **BRANE** selects the configuration that maximizes predicted correctness penalized by cost, exposing a tunable cost-quality tradeoff without retraining. Across MuSiQue, BrowseComp-Plus, and FinanceBench, **BRANE** consistently pushes the cost-quality Pareto frontier, matches the best fixed configuration's accuracy at up to 89% lower cost, and outperforms LLM-routing, rule-based, and fine-tuned Qwen3-4B baselines. These results show that per-query configuration of the full retrieval pipeline is a practical alternative to static workload-level tuning.
Memory-augmented language agents are increasingly deployed in affective applications such as emotional support, where understanding and responding to users' latent emotional needs is critical. However, existing research often treats memory as a tool for factual retrieval, overlooking its role in shaping users' emotional experiences. In this work, we introduce ENPMR-Bench, a benchmark for evaluating Emotional Need-aware Proactive Memory Retrieval (ENPMR), a core capability that enables agents to infer users' latent emotional needs and proactively retrieve appropriate memories to support empathetic interaction. Grounded in Maslow's hierarchy of needs, ENPMR-Bench includes over 1,800 memory-augmented dialogues and defines structured mappings between emotional needs and supportive memory types. Experimental results demonstrate that current retrieval paradigms, including both embedding-based and LLM-driven approaches, exhibit substantial deficiencies, with empathy scores significantly lagging behind golden memory conditions. While chain-of-thought prompting improves the alignment between inferred emotional needs and retrieved memories to some extent, a notable performance gap remains. Together, these findings reveal critical limitations in current agents and outline directions for advancing personalized emotional support through need-sensitive memory retrieval.
展开完整摘要收起摘要↓
Memory-augmented language agents are increasingly deployed in affective applications such as emotional support, where understanding and responding to users' latent emotional needs is critical. However, existing research often treats memory as a tool for factual retrieval, overlooking its role in shaping users' emotional experiences. In this work, we introduce ENPMR-Bench, a benchmark for evaluating Emotional Need-aware Proactive Memory Retrieval (ENPMR), a core capability that enables agents to infer users' latent emotional needs and proactively retrieve appropriate memories to support empathetic interaction. Grounded in Maslow's hierarchy of needs, ENPMR-Bench includes over 1,800 memory-augmented dialogues and defines structured mappings between emotional needs and supportive memory types. Experimental results demonstrate that current retrieval paradigms, including both embedding-based and LLM-driven approaches, exhibit substantial deficiencies, with empathy scores significantly lagging behind golden memory conditions. While chain-of-thought prompting improves the alignment between inferred emotional needs and retrieved memories to some extent, a notable performance gap remains. Together, these findings reveal critical limitations in current agents and outline directions for advancing personalized emotional support through need-sensitive memory retrieval.
作者Haoxuan Jia, Yang Liu, Bin Chong, Yingguang Yang, Yancheng Chen, Jiayu Liang, Qian Li, Hanning Lu, Kefu Xu, Hao Zheng, Chongyang Zhang, Hao Peng, Philip S. Yu
Finance LLM agents must simultaneously block prompt-induced unauthorized actions and approve legitimate multi-step business workflows. However, boundary filters often miss irreversible mid-trajectory tool calls, while post-hoc LLM judges perform auditing only after termination — too late for intervention and at a computational cost that scales linearly with trace length. We present FinHarness, an inline safety harness that wraps a finance agent end-to-end with three components: a Query Monitor that fuses single-turn intent with cross-turn drift, a Tool Monitor that evaluates each prospective tool call, and a Cascade module that integrates per-step risk and adaptively routes verification between a lightweight and an advanced-tier LLM judge. Fired risk factors are re-injected into the agent input as ex-ante evidence, enabling the agent to refuse, re-plan, or approve on its own. On FinVault, routed FinHarness cuts ASR from 38.3% to 15.0% while largely preserving benign approval ($41.1% \to 39.3%$), and uses $4.7\times$ fewer advanced-judge calls than an always-advanced ablation.
展开完整摘要收起摘要↓
Finance LLM agents must simultaneously block prompt-induced unauthorized actions and approve legitimate multi-step business workflows. However, boundary filters often miss irreversible mid-trajectory tool calls, while post-hoc LLM judges perform auditing only after termination — too late for intervention and at a computational cost that scales linearly with trace length. We present FinHarness, an inline safety harness that wraps a finance agent end-to-end with three components: a Query Monitor that fuses single-turn intent with cross-turn drift, a Tool Monitor that evaluates each prospective tool call, and a Cascade module that integrates per-step risk and adaptively routes verification between a lightweight and an advanced-tier LLM judge. Fired risk factors are re-injected into the agent input as ex-ante evidence, enabling the agent to refuse, re-plan, or approve on its own. On FinVault, routed FinHarness cuts ASR from 38.3% to 15.0% while largely preserving benign approval ($41.1% \to 39.3%$), and uses $4.7\times$ fewer advanced-judge calls than an always-advanced ablation.
Vision-Language Models (VLMs) have shown rapid progress in mobile GUI navigation. This paper presents a systematic study of data scaling, benchmarking, and reasoning for VLM-based agents in this domain. To facilitate rigorous evaluation, we introduce HyperTrack, a large-scale dataset with over 16000 real-world tasks across more than 650 Chinese mobile applications, along with GUIEvalKit, an open-source toolkit for unified benchmarking of VLMs on offline GUI navigation tasks. Using HyperTrack, we analyze the effects of training data scale on both supervised and reinforcement-based finetuning. Our results show that reinforcement-based finetuning consistently outperforms supervised finetuning, particularly in out-of-domain settings, highlighting the synergy between data scaling and reinforcement learning. Leveraging GUIEvalKit, we further benchmark state-of-the-art (SOTA) VLMs and analyze how interaction history and reasoning capabilities influence task completion. Together, HyperTrack and GUIEvalKit provide a comprehensive platform for developing and evaluating VLM agents in mobile GUI navigation tasks.
展开完整摘要收起摘要↓
Vision-Language Models (VLMs) have shown rapid progress in mobile GUI navigation. This paper presents a systematic study of data scaling, benchmarking, and reasoning for VLM-based agents in this domain. To facilitate rigorous evaluation, we introduce HyperTrack, a large-scale dataset with over 16000 real-world tasks across more than 650 Chinese mobile applications, along with GUIEvalKit, an open-source toolkit for unified benchmarking of VLMs on offline GUI navigation tasks. Using HyperTrack, we analyze the effects of training data scale on both supervised and reinforcement-based finetuning. Our results show that reinforcement-based finetuning consistently outperforms supervised finetuning, particularly in out-of-domain settings, highlighting the synergy between data scaling and reinforcement learning. Leveraging GUIEvalKit, we further benchmark state-of-the-art (SOTA) VLMs and analyze how interaction history and reasoning capabilities influence task completion. Together, HyperTrack and GUIEvalKit provide a comprehensive platform for developing and evaluating VLM agents in mobile GUI navigation tasks.
A prevalent assumption in LLM agent deployment holds that more structured harnesses universally improve reliability, and that higher-capability models need proportionally less structural guidance — together implying a monotone inverse relationship between model capability tier and optimal harness complexity. We test this hypothesis through a controlled 432-run experiment crossing six models across four capability tiers with three harness conditions (light, balanced, strict) on HEAT-24, a 24-task synthetic benchmark with git-based workspace verification. Our results refute the monotone inverse relationship on two fronts. First, for the frontier chat model evaluated (Gemini 2.5 Flash), increased harness verbosity lowers VTSR by 29-38 percentage points — a harness-complexity paradox. Second, for the frontier reasoning model evaluated (Qwen3.5-122B, extended thinking enabled), strict harness achieves the highest VTSR (91.7%) and the lowest latency, the opposite of the prediction. Within the constrained tier, a 2B model (Gemma4:e2B) matches strong-open-tier stability at 91.7% across all harnesses. Because each tier is represented by a single model in this study, these results should be interpreted as model-specific observations; harness sensitivity appears non-monotone across the models evaluated, and depends critically on model type (chat vs. reasoning). We introduce a six-label failure taxonomy showing that format_violation dominates capable-model failures while wrong_file dominates low-capability failures, and we derive practical tier-aware harness selection guidelines.
展开完整摘要收起摘要↓
A prevalent assumption in LLM agent deployment holds that more structured harnesses universally improve reliability, and that higher-capability models need proportionally less structural guidance — together implying a monotone inverse relationship between model capability tier and optimal harness complexity. We test this hypothesis through a controlled 432-run experiment crossing six models across four capability tiers with three harness conditions (light, balanced, strict) on HEAT-24, a 24-task synthetic benchmark with git-based workspace verification. Our results refute the monotone inverse relationship on two fronts. First, for the frontier chat model evaluated (Gemini 2.5 Flash), increased harness verbosity lowers VTSR by 29-38 percentage points — a harness-complexity paradox. Second, for the frontier reasoning model evaluated (Qwen3.5-122B, extended thinking enabled), strict harness achieves the highest VTSR (91.7%) and the lowest latency, the opposite of the prediction. Within the constrained tier, a 2B model (Gemma4:e2B) matches strong-open-tier stability at 91.7% across all harnesses. Because each tier is represented by a single model in this study, these results should be interpreted as model-specific observations; harness sensitivity appears non-monotone across the models evaluated, and depends critically on model type (chat vs. reasoning). We introduce a six-label failure taxonomy showing that format_violation dominates capable-model failures while wrong_file dominates low-capability failures, and we derive practical tier-aware harness selection guidelines.
Background: Large language models are typically evaluated as models, benchmarks, or short conversational episodes. Less is known about what happens when an agent is embedded persistently in a real academic research environment with durable memory, local files, external tools, scheduled routines, delegated roles, and explicit safety protocols. Methods: A structured self-observed implementation case study was conducted from January 31 to May 25, 2026. The unit of analysis was the persistent human-agent environment: researcher, agent runtime, memory layer, tools, repositories, scheduled jobs, specialized agent roles, and governance rules. Outcomes were organized using PARE-M (Persistent Agentic Research Environment Measurement), a measurement framework covering architecture, utilization, artifact production, resource use, reproducibility, and governance. Results: Recoverable main-agent telemetry contained 75,671 de-duplicated records across 96 active days, with 8,059 user-role and 23,710 assistant-role messages. The workspace included 502 memory-related files, 17 configured agent directories, and 57 skill files. Active system time was 579.7 hours (30-minute capped-gap estimate). Memory-derived records identified 482 output-proxy events and 889 failure, verification, correction, or protocol-proxy events. A strict May 2026 trajectory subset captured 627 model-completed events and 73.95 million recorded tokens, of which 82.9% were cache reads. Conclusions: The workflow was cache-dominant, suggesting that persistent agentic environments may shift the economic unit from cost per token to cost per completed artifact. Future evaluations should use artifact-level denominators, reproducible parsing rules, correction taxonomies, and independent coding of governance events.
展开完整摘要收起摘要↓
Background: Large language models are typically evaluated as models, benchmarks, or short conversational episodes. Less is known about what happens when an agent is embedded persistently in a real academic research environment with durable memory, local files, external tools, scheduled routines, delegated roles, and explicit safety protocols. Methods: A structured self-observed implementation case study was conducted from January 31 to May 25, 2026. The unit of analysis was the persistent human-agent environment: researcher, agent runtime, memory layer, tools, repositories, scheduled jobs, specialized agent roles, and governance rules. Outcomes were organized using PARE-M (Persistent Agentic Research Environment Measurement), a measurement framework covering architecture, utilization, artifact production, resource use, reproducibility, and governance. Results: Recoverable main-agent telemetry contained 75,671 de-duplicated records across 96 active days, with 8,059 user-role and 23,710 assistant-role messages. The workspace included 502 memory-related files, 17 configured agent directories, and 57 skill files. Active system time was 579.7 hours (30-minute capped-gap estimate). Memory-derived records identified 482 output-proxy events and 889 failure, verification, correction, or protocol-proxy events. A strict May 2026 trajectory subset captured 627 model-completed events and 73.95 million recorded tokens, of which 82.9% were cache reads. Conclusions: The workflow was cache-dominant, suggesting that persistent agentic environments may shift the economic unit from cost per token to cost per completed artifact. Future evaluations should use artifact-level denominators, reproducible parsing rules, correction taxonomies, and independent coding of governance events.
LLM-generated peer reviews are increasingly common at major venues, yet their deficiencies are hard to detect because they are uniformly fluent and well-structured. Existing work either classifies authorship without judging quality, or scores quality with features designed for human-written reviews; no prior system detects deficiencies in LLM-generated reviews at the level of individual defect types. To bridge the gap, we introduce TADDLE, a Tool-Augmented Agent for Detecting Deficient LLM-Generated Peer Reviews, together with the first expert-annotated benchmark for this task. Our benchmark comprises 1,800 reviews on 50 ICLR 2025 papers, multi-label-annotated by 18 domain experts against a taxonomy of six defect categories (plus a non-deficient label). TADDLE decomposes detection into four specialized analysis tools — Verify, Correct, Complete, and Transform — orchestrated by an agent; an integrator synthesizes their outputs into binary and multi-label classifications via two-stage semi-supervised learning. Extensive experiments show that TADDLE performs strongly on both binary detection and the multi-label classification task. We release the benchmark and code at https://github.com/AquariusAQ/TADDLE.
展开完整摘要收起摘要↓
LLM-generated peer reviews are increasingly common at major venues, yet their deficiencies are hard to detect because they are uniformly fluent and well-structured. Existing work either classifies authorship without judging quality, or scores quality with features designed for human-written reviews; no prior system detects deficiencies in LLM-generated reviews at the level of individual defect types. To bridge the gap, we introduce TADDLE, a Tool-Augmented Agent for Detecting Deficient LLM-Generated Peer Reviews, together with the first expert-annotated benchmark for this task. Our benchmark comprises 1,800 reviews on 50 ICLR 2025 papers, multi-label-annotated by 18 domain experts against a taxonomy of six defect categories (plus a non-deficient label). TADDLE decomposes detection into four specialized analysis tools — Verify, Correct, Complete, and Transform — orchestrated by an agent; an integrator synthesizes their outputs into binary and multi-label classifications via two-stage semi-supervised learning. Extensive experiments show that TADDLE performs strongly on both binary detection and the multi-label classification task. We release the benchmark and code at https://github.com/AquariusAQ/TADDLE.
Mobile graphical user interface (GUI) agents enable AI models to autonomously operate smartphones on behalf of users. However, most existing systems focus primarily on optimizing task accuracy and rely on cloud-hosted models for inference, which introduces privacy concerns and network-dependent latency. As a result, fully on-device deployment of mobile GUI agents remains underexplored. We propose MobileExplorer, a new framework that accelerates on-device inference for vision-based mobile GUI agents via online exploration. The key idea is to exploit the long per-step reasoning time of vision-language models (VLMs) by performing lightweight, parallel exploration of UI elements. During model inference, the agent proactively probes semantically relevant UI elements and records these exploration traces as structured memory. To ensure reliable execution in live mobile environments, we design a two-level rollback mechanism that robustly restores the initial UI state when a fast but naive backtracking strategy fails. The collected exploration traces are then summarized into concise contextual hints and injected into the prompt to enhance the subsequent reasoning step. We evaluate MobileExplorer on multiple off-the-shelf devices using the AndroidWorld benchmark, as well as newly designed, more complex tasks and dynamic on-device environments. MobileExplorer reduces the average number of reasoning steps and end-to-end latency by 23%, while maintaining or improving task success rates by up to 5%. A video demonstration of MobileExplorer performance in the real world is available at https://youtu.be/thK7MJmdlvM .
展开完整摘要收起摘要↓
Mobile graphical user interface (GUI) agents enable AI models to autonomously operate smartphones on behalf of users. However, most existing systems focus primarily on optimizing task accuracy and rely on cloud-hosted models for inference, which introduces privacy concerns and network-dependent latency. As a result, fully on-device deployment of mobile GUI agents remains underexplored. We propose MobileExplorer, a new framework that accelerates on-device inference for vision-based mobile GUI agents via online exploration. The key idea is to exploit the long per-step reasoning time of vision-language models (VLMs) by performing lightweight, parallel exploration of UI elements. During model inference, the agent proactively probes semantically relevant UI elements and records these exploration traces as structured memory. To ensure reliable execution in live mobile environments, we design a two-level rollback mechanism that robustly restores the initial UI state when a fast but naive backtracking strategy fails. The collected exploration traces are then summarized into concise contextual hints and injected into the prompt to enhance the subsequent reasoning step. We evaluate MobileExplorer on multiple off-the-shelf devices using the AndroidWorld benchmark, as well as newly designed, more complex tasks and dynamic on-device environments. MobileExplorer reduces the average number of reasoning steps and end-to-end latency by 23%, while maintaining or improving task success rates by up to 5%. A video demonstration of MobileExplorer performance in the real world is available at https://youtu.be/thK7MJmdlvM .