每天知道
你的领域
出了什么

Modern text-to-image (T2I) systems can be improved without modifying generator parameters by adapting the external system around frozen generators. However, existing approaches typically optimize a predefined dimension, such as prompts, routing, or workflows, restricting the space in which generation failures can be corrected. Allowing multiple generator-external responsibilities to evolve provides a broader adaptation space, but introduces a new challenge: visual feedback reveals what failed, but not where persistent evolution should occur or how this space should be explored efficiently. We introduce EvoGen-Harness, a generator-agnostic framework for multi-responsibility image-generation harness evolution, together with Trace (Trajectory-Relative Attribution and Coordinated Evolution). Trace aggregates evidence across stochastic executions, uses failure attribution as a search prior to focus candidate updates, and progressively re-attributes residual failures to coordinate evolution across responsibilities, while No-Patch and held-out validation prevent unnecessary or harmful updates. Across GenEval2, T2I-CompBench++, and WISE, EvoGen-Harness improves over the strongest evaluated baselines by +0.2633, +0.0720, and +0.0752, respectively, while achieving 87.9-91.4% attribution recall, 94.8% No-Patch accuracy, and only 1.9% regression. These results demonstrate that attribution-guided multi-responsibility evolution can substantially enhance frozen T2I systems beyond single-dimension adaptation.

Recent video-based world models pair the scalability of autoregressive (AR) prediction with the visual quality of diffusion models. The choice of scene tokenizer is paramount for the optimal performance of each of these, both in terms of fidelity and semantics. Flexible-length, coarse-to-fine tokenizers yield exactly that: the first coarse tokens carry the clip's global semantics while later tokens further specify details. Existing flexible tokenizers only apply a representation-alignment (REPA) loss on early decoder hidden states, a target the decoder can partly meet from its noised input instead. We introduce SemanTok, a flexible video tokenizer that feeds frozen DINO features into its encoder and adds lightweight heads that reconstruct them from each retained token prefix alone. SemanTok achieves high semantic alignment and video fidelity at every AR model size: a 201M SemanTok AR model matches or beats a VideoFlexTok AR model $3.4\times$ its size, and larger SemanTok AR models further improve fidelity. It keeps semantic alignment on out-of-distribution classes and gives the decoder higher semantic alignment at every noise level, including pure noise. It performs well in both reconstruction and generation, and its short token prefixes are cheaper to predict and give better generation fidelity, with pixel detail deferred to later tokens.

We present PLACE, a method that extends the pretrained any-to-audio model AudioX for binaural generation from arbitrary combinations of text, video, and optional audio prompts. PLACE augments video conditioning with Perception Encoder Core features, aligns text and video representations to derive spatial cues, and applies a conditioning-dependent low-rank transformation to the generated latent. The adapter is supervised via decoded-audio interaural level and time difference objectives. Trained on MRSAudio, PLACE improves most metrics over ViSAGe on FAIR-Play and achieves an improved SpatialCLAP score over SpatialSonic on the BEWO-1M Single Static test split. Listener evaluations favor PLACE for video-to-audio and out-of-distribution text-to-audio generation, demonstrating flexible multimodal control and improved spatial consistency.

Personalized image generation has remained narrowly focused on conditional synthesis from curated visual exemplars, rather than capturing who a user is. In practice, however, a user's personal context is much richer, comprising reviews, posts, images, captions, and metadata accumulated over time. A truly personalized generator should leverage this history to produce images aligned with the user's lifestyle and aesthetic preferences. To this end, we introduce the first unified benchmark for personalized image generation from user histories. The benchmark comprises two complementary tasks and a multi-axis evaluation protocol that assesses target fidelity, visual quality, user distinguishability, semantic alignment with the user's history, and task-specific utility. Grounded in real-world e-commerce and social media settings, the benchmark includes: (1) Personalized Scene Generation, which places a given object in a scene that reflects a user's preferences and lifestyle, motivated by personalized product presentation; and (2) Personalized Creative Generation, which generates a novel image on a specified topic that is faithful to a user's aesthetic and visual identity, motivated by social media content creation. We further propose PEARL, which couples a multimodal reasoner with a frozen image generator in an interleaved reason-reflect loop optimized with differential data reward. Across both tasks, PEARL outperforms strong baselines, achieving an average improvement of 15% across personalization metrics.

One-step generative models construct a static generator through iterative training-time transport. Existing transport objectives primarily assess distributional motion, although a neural generator needs to realize the requested sample displacements jointly through shared parameter updates. The training-time construction raises the question: once training becomes the iterative process that constructs the final one-step map, what to optimize: the next distributional move, or the route by which the finite generator learns the final map? To address the question, we introduce Training Dynamics Action (TDAction), which selects transport targets according to local shared-parameter realization cost while retaining a prescribed level of distributional progress. We formulate the cost as a soft-terminal control problem and derive a closed-form Batch Tangent Action-to-Go value that accounts for parameter effort and terminal mismatch. The criterion captures cross-sample interactions omitted by independent pairwise costs; under isotropic mobility, the criterion agrees with quadratic Euclidean assignment for deterministic balanced couplings. Randomized tangent probes provide a low-rank implementation that constructs shared detached targets without adding an inference-time trajectory. Controlled studies examine the relationship between generator geometry, transport selection, and realized local action. On ImageNet $256\times256$, TDAction attains an FID below $1.1$ without distillation.

We introduce Grand Canonical Generators (GCG), a generative framework that extends Boltzmann generators to the grand canonical ensemble. We present two designs. The first conditions a variable-size generative model on the chemical potential, sampling particle number and configuration jointly. The second factorizes the grand canonical distribution into a particle-number distribution and the corresponding canonical Boltzmann density. This factorized formulation can use any existing Boltzmann generator for the canonical component, encodes the known linear chemical-potential dependence analytically, and yields a tractable likelihood that supports self-normalized importance sampling (SNIS). Empirically, GCG accurately reproduces grand canonical observables on a Lennard--Jones fluid and methane adsorption in a zeolite, demonstrating generalization across chemical potentials and correction via SNIS and grand canonical Monte Carlo.

Preference-based alignment methods typically optimize against a single preference model, and can therefore be brittle when pairwise preferences are uncertain: noisy, heterogeneous, or shift after deployment. To address these issues, we propose Robust Nash Alignment, a game-theoretic framework for alignment to uncertain pairwise preferences. Our formulation has a major learner seeking a policy with a large worst-case win rate against both an adversarial competitor and any preference kernel lying in an ambiguity set around a nominal preference. When the ambiguity set captures the uncertainty in preferences, the resulting robust objective of the game directly yields a certified lower bound on worst-case performance. However, we note this problem is computationally challenging to optimize, and to address this, we introduce a four-player primal-dual proxy game involving the leader policy, follower policy, adversarial kernel, and dual variable, and develop a single-loop optimistic mirror descent-ascent algorithm for it. We show that the proxy always lower-bounds the truncated hard-constrained objective, quantify the proxy-to-hard gap, and characterize an exactness condition under which the proxy recovers the robust objective. We then prove an \(\mathcal{O}(1/\sqrt{T})\) average-iteration convergence for the proxy-game duality gap, which implies a near-optimal robust policy for the original robust objective. Experiments on controlled tabular games and LLM alignment with uncertain preference further validate the convergence theory and show improved performance over nominal baselines.

Editing a chest radiograph requires completing the requested change while preserving unrelated content. We study a latent diffusion editor with an instruction-independent source trajectory and an instruction-conditioned editing trajectory. A learned gate mixes their post-sampler candidates at each executed step, and a separate image-space mask composites the decoded proposal with the source. On 2,400 MIMIC-derived requests, 2,244 outputs met the joint target, preservation, quality, and coverage rubric (93.5%), and target completion was 97.4%. Joint validity exceeded a matched composition-only control by 1.7 percentage points (paired patient-cluster 95% interval, 0.6--2.8). At a fixed learned mask, the learned-gate proposal improved joint validity by 1.6 points (0.8--2.4); at a fixed learned proposal, the learned mask improved it by 2.6 points (1.7--3.5). Across three training seeds, mean joint validity was 93.5% with a 0.3-point sample standard deviation. A blinded 240-request assessment yielded adjudicated joint validity of 93.3% for the full editor and 91.7% for composition only. Protected-region mean absolute error decreased from 0.0190 in the raw proposal to 0.0075 after composition. These findings distinguish recurrent-gating effects on the proposal from preservation through final composition in the assessed cohort.

Recent advances in distillation and flow-map models have enabled deterministic one- or few-step generation for high-quality data, facilitating a new branch of reward alignment approaches that directly optimize the initial noise from a Gaussian distribution. However, most existing initial-noise optimization methods rely on first-order gradient information, which is either inapplicable or suffers from instability and inefficiency in black-box reward scenarios. Here, we introduce ZeNOVA, a stable and efficient initial noise alignment method in a gradient-free manner. Specifically, we address existing algorithms' major challenge in black-box scenarios through annealed soft-value guidance, manifold-constrained hyperspherical Langevin dynamics, and Metropolis-Hastings jumping. Extensive experiments on image and video generative models show that ZeNOVA outperforms all evaluated zeroth-order baselines by optimizing the initial noise toward higher rewards substantially more stably while exploiting the geometry of the Gaussian prior, demonstrating its practical applicability to various black-box reward alignment.

Recent joint audio-video generative models can synthesize realistic videos with synchronized sound, but typically generate audio as a single mixed track. This limits source-level control and differs from practical audiovisual workflows, where speech, music, sound effects, and ambient sounds are represented as separate editable tracks. We introduce Soundwich, a training-free framework that transforms a frozen joint audio-video flow-matching model into a generator of multiple synchronized, independently editable audio stems coupled to a shared video. Soundwich generates separate audio stems with explicit control over their temporal activity. To keep separately generated sounds coherent, we introduce a shared scene representation that communicates global audiovisual context across stems while preserving their source-level separation. We further route cross-modal interactions between each audio stem and its corresponding visual source, improving audiovisual consistency. The resulting stems remain synchronized with the video and can be independently retimed, muted, replaced, or remixed. Experiments and human evaluations show improved temporal control, source separation, and naturalness, while enabling flexible source-level editing within coherent audiovisual generation. Code is available at https://github.com/CodyNing/Soundwich.

Identifying which speaker a listener is attending to in a noisy room — the cocktail-party problem — is the missing ingredient for next-generation hearing aids and brain-computer interfaces: it tells the device whose voice to amplify. Auditory attention decoding (AAD) reads this answer from EEG, but the literature splits into disconnected pieces: directional-AAD classifies side but does not map side to stream; regression-based source-AAD ranks candidate streams by a single Pearson correlation that is intrinsically noisy at the 1-5 s windows real devices need; and envelope reconstruction has no native AAD rule. We argue the right object is not any single statistic but the conditional likelihood of the attended envelope given EEG, and we make this practical with NEUROTOKEN: a single network whose three heads share one EEG front-end, with a conditional flow-matching head (ATTUNEFLOW) that scores candidates by an integrated velocity-residual likelihood ratio. Two inference-time ensembles — QUADTRACK (four complementary statistics) and ENV-FLOW (z-normalised QUADTRACK+ATTUNEFLOW) — absorb per-statistic failure modes for free. On KU Leuven, DTU, and NJU at 5 s, ATTUNEFLOW lifts per-segment source-AAD by 9%-16% over the strongest non-generative baseline and shrinks across-subject variance by ~3x; trial-level fusion exceeds 93% on two of three datasets. In parallel reproductions we show that canonical 95-97% direction-AAD numbers collapse by 17%-45% under a strict trial-disjoint protocol, clarifying both the true ceiling and why a likelihood-based formulation is needed.

Flow matching (FM) learns generative transport by fitting continuous-time motion from a simple source distribution to the data distribution. Most existing methods use first-order bridges: once a source and a target sample are paired, the path is a straight motion with constant velocity. FM with optimal transport (OT) improves the pairing, but the bridge itself remains linear, limiting its ability to model curved motion, acceleration, and changing directions. A natural remedy is to use second-order phase-space dynamics; however, learning the bridge requires target-side terminal-velocity information that static datasets do not provide. We propose Variational Terminal-Velocity Flow Matching (VTV-FM), a second-order FM framework that derives the missing velocity by minimizing acceleration energy, yielding a closed-form closure for static data. The same minimum-acceleration variational construction also defines the OT pairing cost and the acceleration targets used for training. Experiments on low-dimensional datasets, PDE-governed physical fields, and CIFAR-10 show that VTV-FM improves transport geometry and generation quality over first-order and high-order FM baselines.

Representation alignment has become an effective way to accelerate diffusion training, but its benefits do not transfer reliably to pixel-space clean-image prediction. In JiT, we find that auxiliary feature alignment can improve access to semantic features while reducing access to image variation needed for clean-image prediction, creating a mismatch between the auxiliary objective and the denoising task. This suggests a different principle: auxiliary supervision should improve the prediction target itself rather than impose a separate representation target. We introduce JAx (Just Align x), a prediction-supervision method that aligns clean-image predictions across noise levels. JAx couples a noisier student observation with a cleaner observation through a Markov degradation that preserves the original JiT input distribution. Under this coupling, the oracle prediction from the cleaner state has the same conditional mean as the optimal JiT target, while its conditional target covariance is no greater. Thus, oracle prediction alignment preserves the population JiT objective up to a constant while providing a lower-variance training target. To make this construction practical with an imperfect EMA teacher, JAx combines ground-truth supervision with a reliability-gated coupling band that selects nearby teacher states based on prediction risk. On ImageNet 256x256, JAx consistently improves FID and accelerates convergence across JiT-B/16, L/16, and H/16, without an external encoder or changes to the architecture or sampling procedure. Gradient diagnostics further show reduced minibatch gradient variance, while ablations demonstrate that the gains cannot be explained by time reweighting alone. These results show that prediction-space supervision provides a simple and principled alternative to representation alignment for pixel-space generative models.

Diffusion models with prompt and reference image-guided editing have seen rapid progress, yet they remain too coarse for pixel-level control. One promising direction is to incorporate a soft mask that specifies spatially varying edit strengths but training such fine-grained control demands expensive pixel-wise annotations, while existing zero-shot methods often yield unsatisfactory results. We introduce SoftPaint, a new zero-shot sampling method that leverages soft masks to enable a continuous spectrum of edits, from fully preserving the original content to completely re-synthesizing the masked region. Going beyond zero-shot inpainting methods, we design a Langevin-iteration-based sampler that respects per-pixel soft mask strengths, which applies universally to image and video diffusion models, enabling tasks such as video editing. The method is gradient-free, memory-efficient, and achieves smooth, pixel-level edits across multiple image and video backbones.

Inference-time steering enables pretrained diffusion models to satisfy new constraints without full retraining. However, specificity-aware generation is difficult: repelling samples from a negative reference distribution can also erode the positive distribution where the two overlap. The key challenge is to suppress negative mass while minimally distorting the positive distribution. We address this problem by formulating specificity-aware steering as a target-design problem and deriving a target distribution from an overlap-based objective. The resulting target keeps the desired reference distribution only in regions where it is sufficiently preferred over the undesired reference distribution, giving a likelihood-ratio interpretation of specificity. To sample from the corresponding time-dependent target path, we develop a Sequential Monte Carlo sampler with a variance-minimized local proposal. We further introduce a practical fixed-noise optimization procedure with the Jacobian--vector products with the desired and undesired score fields. Experiments on synthetic task, class-contrastive generation, text-to-image tasks and peptide-MHC (p-MHC) binder show that the proposed method suppresses undesired regions more effectively, reduces mode shift, and improves sampling stability by decreasing the SMC weight collapse compared with negative-guidance baselines. Code is available at: https://github.com/WangLuran/Specificity-Aware-Diffusion-Steering

A persistent agent must decide both what to retain as information arrives and what to surface once a query appears, yet memory evaluations can confound these decisions by comparing methods that differ in both retention and selection. We build a streaming-recall benchmark crossing retention and selection rules and evaluate every condition on the same 300 seeded episodes. Holding access fixed, query-aware selection improves required-fact recall by 15.5 percentage points (95% CI: 12.8 to 18.2), whereas a mixed comparison that also changes history access reports a 68.7-point advantage, of which 53.2 points are attributable to access. Under bounded retention, query-aware, dense, and oracle selection reach the retention ceiling, and all 319 observed failures in the bounded recency condition are caused by eviction rather than ranking errors. Recall falls to 0% as targets recede sufficiently far into the past. Repeating the evaluation on SQuAD preserves the retention ceiling while showing that dense retrieval can outperform lexical retrieval on natural text. These results show that bounded-memory evaluations should hold access fixed and report retention and selection separately.

We study how AI agents discover and validate scientific formulas using a controlled case study of the hydrotope, a recently discovered geometric formula that combines the different polynomial pieces of nonlinear surface-wave scattering into one global expression. This problem is deceptively difficult: simple formulas can hold within individual frequency regions, but the global result must identify their boundaries and combine exponentially many potentially active terms. We reconstruct how the formula was originally discovered through human--agent collaboration and analyze 18 single-prompt rediscovery runs under no hint and two forms of human guidance: a false hint representing an incorrect prior and a true hint representing domain-informed insight. Only four recover the formula across all kinematic chambers (i.e., regions in which a single polynomial form applies), while most unsuccessful runs find correct chamber polynomials but fail to combine them or test their full domain. Conventional and LLM-assisted symbolic regression and standard machine-learning regressors likewise fail to recover the global formula in our experiments. Guided by these failure modes, we test a PI$+$two-student workflow in which a coordinating lead agent assigns complementary analytic and numerical tasks to two research agents and independently evaluates their results. The PI$+$two-student team successfully rediscovers the complete hydrotope formula, while the same workflow applied to the harder three negative wavenumber problem discovers a new independent verified analytic expression for the six-point amplitude $A_6$.

Large language model agents rely on external harnesses to pass information between the model and its environment and to recover from execution errors. Yet recovery is usually judged only by average task success. This hides an important tension. The same operation can rescue a failing trajectory or disrupt one that would otherwise succeed. We frame recovery as a causal decision problem. Starting from the same execution state, we compare what happens with and without recovery, separate rescue from harm, and study how the value of recovery changes over time. We then introduce the Causal Intervention Router (CIR), a lightweight policy that uses information available before recovery to decide when intervention is worthwhile. On long-horizon ALFWorld tasks with Qwen3-14B, CIR raises success from 70.33% to 73.33%, a gain of 3.00 percentage points. It leaves all evaluated trajectories with correct observations untouched. Additional controls show that the benefit of recovery cannot be explained solely by the new observation returned by the environment. These results provide a practical way to evaluate recovery and apply it selectively.

Natural-language skills are textual procedural memories through which large language model (LLM) agents retain reusable task knowledge without updating model weights. Existing methods typically treat skills as either static artifacts or monolithic documents optimized using aggregate validation scores as feedback. However, representing a skill as a monolithic document restricts optimization to its textual content, without explicitly modeling the structure through which procedural knowledge is retrieved and executed. We identify a key distinction between learning what knowledge to retain and determining how to organize it: textual updates should first be validated through execution evidence, after which the retained knowledge should be structured according to its procedural dependencies and retrieval requirements. To this end, we introduce SkillSpec, a two-phase framework comprising consensus-gated evolution and representation specialization. In the consensus-gated phase, complementary editing intents generate complete candidate skills. An update is committed only when paired evaluations reach consensus, requiring sufficient overall improvement and non-negative aggregate paired gain in every repeated evaluation. In the specialization phase, signals of process and redundancy sensitivity derived from the full optimization trajectory, including accepted and rejected candidates, guide the selection of a flat, graph, or hybrid representation.Across six benchmarks and three target language models, SkillSpec improves average success rate over SkillOpt by 6.89%, averaged across the three models. These results demonstrate that reliable skill evolution and representation specialization address complementary objectives: deciding what knowledge to retain and how to structure it for inference.

LLM agents generate intermediate reasoning and actions token by token, making extended interactions slow and computationally expensive. Jev-style models offer fast probabilistic predictions over finite fields, but require those fields to be specified in advance. This requirement limits autonomous task solving, where the available actions must be derived from natural language instructions and adapted through interaction. We introduce JevSpawn, a compositional policy that connects natural language task specifications to finite probabilistic exploration. Parallel action spawning is coupled with feedback driven branch selection, representation revision, and recovery from retained alternatives. Shared action structure and model prefixes reduce repeated generation and context computation without additional training. Evaluations on eight benchmark tasks against seven agent baselines and a TypeSafe Jev variant establish JevSpawn as a promising approach to structured agentic inference, with improved task performance and faster navigation.

Multi-agent systems derive their capabilities from sharing evidence, delegating tasks, and combining information across agents. The same process creates a safety problem: contributions that are admissible in isolation can jointly enable a prohibited use. Blocking every sensitive action avoids disclosure but defeats the purpose of collaboration. We introduce authorization-paired evaluation, which makes blocking prohibited uses and completing required authorized uses a joint success criterion, and FlowReview, a framework connecting object resolution, permission ranking, and deterministic enforcement. In controlled composition experiments, reviewing combined artifacts reduces the denied-commit rate from 86.0% to zero with no loss of authorized supply. Our findings show that preserving information and lineage alone does not ensure correct permission attribution. Object identity and permission must remain connected to execution through components whose outputs can be verified. Together, these findings establish a system-level requirement for multi-agent safety: govern composed information flows while preserving the authorized capabilities that make collaboration useful.

The integration of Large Language Models (LLMs) into multi-agent systems has shown great potential for automated debugging. Yet nearly all current frameworks rely on rigid, predefined architectures: the number of agents, their roles, and their interaction patterns are fixed before any analysis of the bug occurs. This one-size-fits-all approach is fundamentally mismatched to the heterogeneous nature of software defects. Simple bugs waste resources on unnecessary coordination, while complex ones suffer from insufficient or poorly aligned expertise. This paper introduces ASAD, an adaptive agentic system for debugging that configures its team according to the nature and complexity of each bug. ASAD initiates the debugging process by analyzing the faulty code and dynamically determines the number of agents to deploy, the specialized roles they should have, and the collaboration strategy they should follow. A central coordinator orchestrates this process through iterative planning, reflection, and execution; applying fast single-pass repairs for simple issues while assembling purpose-built teams to tackle more complex failures. We evaluate ASAD on three established benchmarks: Defects4J, DebugBench, and CodeFlaws, using multiple LLMs, such as DeepSeek-V3, Qwen-3 and GPT-5. ASAD consistently improves bug-fix rates by 12--20% over chain-of-thought(CoT) prompting and consistently outperforms static multi-agent systems by 4--9% in fix precision while reducing average agent usage by 32%. Crucially, our system dynamically adjusts the number and roles of agents: it resolves simple bugs with minimal coordination and scales agent involvement only for more complex cases.

An autonomous cyber defender trained with reinforcement learning (RL) is typically tied to the network on which it was trained, limiting its ability to generalize as network scale changes. Hierarchical RL reduces decision complexity by separating strategic targeting from tactical execution, but it does not eliminate this retraining dependence. We investigate whether frozen, zero-shot large language models (LLMs) can provide retraining-free control in hierarchical cyber defense and how performance changes as LLM control is extended from planning to execution. We formulate a controller-agnostic planner-executor hierarchy in which the planner selects a subnet to defend over a fixed horizon and the executor selects defensive actions within that subnet. Using the high fidelity Cyberwheel environment, with its built-in automated red team agent mapped to the MITRE ATT&CK framework, we compare RL+RL, LLM+RL, and LLM+LLM configurations using six models ranging from 3B to 70B parameters, including two cybersecurity-specialized models, across small, medium, and large networks. Replacing only the planner with an LLM yields limited gains as network size increases. In contrast, extending LLM control to execution produces notable improvements for sufficiently capable models. For instance, a frozen general purpose 70B model holds successful lateral movement to approximately 1% of steps and attacker impact near zero across all three network scales using the same model weights, while the RL baseline is retrained for each scale. Our results show that sufficiently capable frozen LLMs can maintain strong defensive performance across the evaluated network scales without task-specific retraining, while also indicating that strong tactical execution is important to realizing the benefits of LLM-based control.

Evaluating AI agents on scientific research tasks is constrained by the time and resources required for the underlying experiments or calculations. In computational materials research, repeating the same expensive simulations across agents and trials can make evaluation impractical. We introduce CompMat-Bench, a benchmark of 94 tasks derived from recently published computational materials studies, each asking agents to complete a step toward achieving the study's scientific goal. We reproduce the research steps in advance and assess agents on preparing inputs and analyzing outputs for expensive simulations, so expensive simulations can be avoided during evaluation. The reproduced inputs and results serve as ground truth for grading agents with fixed rules, without an LLM judge. The benchmark supports four evaluation conditions: single tasks and workflows composed of related tasks, each with full or reduced methodological guidance. With full guidance on single tasks, agents based on three LLMs demonstrate the ability to complete individual materials research steps, with pass rates of 66.0-90.4% across 94 tasks. Both longer workflows and reduced guidance can limit agent performance, but in different ways for different agents: they lower the pass rates of the weaker agents, whereas the strongest agent falls only when a long workflow is combined with reduced guidance. Failure analysis attributes most failures to scientific errors rather than to errors in software usage. CompMat-Bench provides a basis for comparing agents on the steps of real materials research and for analyzing agent failure modes.

As large language model (LLM) agents become widely adopted, they are increasingly deployed for tasks that require persistent monitoring or recurring actions (e.g., market analysis). These agents are expected to operate unattended for days or weeks, act at the right timing, and adapt to the dynamic environment over time. These challenges are not fully captured in the existing long-horizon agent work, as they often consider a static environment that is not temporally changing. We introduce ReLiveGym, a diagnostic evaluation environment of long-lived tasks in which agents act sparsely over simulated weeks of chronologically replayed real-world news, market, and social-media streams. The tasks span diverse levels of time sensitivity, reasoning intensity, and recurrence. Across eight base language models, we investigate how model choice and harness design affect agent performance on such long-lived tasks. Our results show that how agents determine when to act arises as an important harness-design axis for long-lived tasks; and that the optimal design varies across tasks and sometimes model choices as well. We also evaluate how continuous learning from hindsight feedback affects performance and addresses failure modes observed in these long-lived tasks. These findings indicate model choice, action timing mechanism, and use of feedback as important considerations in the design of long-lived agents. Code: https://github.com/SaharaLabsAI/ReLiveGym

Legal research is a core and time-consuming legal workflow. Lawyers must identify controlling authority, verify that it remains valid, reconcile statutes and cases, and synthesize a grounded answer. Language model agents are a natural fit for this retrieval-intensive workflow, and automating even part of it would be valuable. But that value depends on reliability: a single missing authority, stale citation, or wrong legal conclusion can make an otherwise plausible answer unusable. We introduce Legal Research Bench (LRB), a benchmark of 413 open-ended U.S. legal research questions written by experts, each paired with a gold answer, supporting authorities, and a binary grading rubric. We evaluate thirteen frontier models in a harness with web search, case-law search, page parsing, and retrieval tools. We score agent responses through all-pass grading with source verification, where a response is correct only if every required criterion is satisfied and its cited authorities verify. We also validate the LLM judge against expert attorneys ensuring that benchmark scores track attorney judgment. Agents remain far from reliable: among the models we tested, the strongest, Claude Opus 4.8, is fully correct on 42.9% of questions. Performance also varies substantially by task setting: all-pass rates differ across areas of law and are lower on questions requiring reconciliation of conflicting authorities. Across models, more turns, tool calls, and inference cost do not predict higher accuracy.

Autonomous large language model (LLM) agents increasingly interact in network environments where adversarial content can propagate between agents. Known attacks include agent worms, which spread through self-replicating prompt injections or configuration compromises. We introduce memetic trojans, a distinct class of network-mediated attack that exploits agents' tendencies to retransmit and amplify content. Unlike agent worms, whose propagation is adversarially induced, memetic trojans exploit endogenous transmission by embedding adversarial payloads in social contagions: content agents have internal reasons to share. As part of our work, we extract social contagions from Moltbook, a social media platform for LLM agents. Controlled transmission experiments reveal large differences in virality: the most effective contagion is retransmitted in approximately 50% of subsequent agent posts and upvoted at 2.5x the average post's rate. Its memetic trojan counterpart largely inherits these properties. Monte Carlo attack simulations show that memetic trojans amplify expected exposure by up to 3.19x. Network structure and amplification mechanisms strongly shape propagation, producing heavy-tailed outcomes with near network-wide exposure. These results identify endogenous social transmission as a distinct security vulnerability in multi-agent systems. Because propagation does not require agents to follow malicious retransmission instructions, defenses focused on prompt-injection detection or preventing agent compromise cannot alone prevent memetic trojan propagation. Securing large-scale agent ecosystems may require network-level defenses that account for how agent preferences, recommendation mechanisms, and network topology amplify adversarial payloads.

People are increasingly delegating tasks to AI agents, and those agents are increasingly encountering other people's agents over shared resources such as a codebase, a calendar, or a budget. When each agent acts for a different user with different goals, coordination often fails, and the group ends up worse off than if a single agent had acted for everyone. We study this multi-user, multi-agent setting across five frontier models and 77 scenarios in four environments: an API key environment in which agents share a compute budget, a clinic in which they share a calendar, a personal assistant environment in which they share a group order or booking, and a merge queue in which they share a release cutoff. In each scenario, we compare a single agent that serves every user (a coordinator) to a team in which each agent serves one user, with and without a communication channel between the agents. Teams deliver worse group outcomes than the coordinator in every environment: without a channel, they completely collapse in two environments, and even with one, coordination overhead creates substantial gaps. For example, in the personal assistant environment, the coordinator fulfills a targeted user request about twice as often as teams. We identify distinct behaviors associated with this poor group-level performance, including stalling as teams grow, overriding each other's actions, and fabricating claims. We find effective but environment-specific mitigations, such as a team lead, explicit procedural instructions, and a platform check that makes an agent read its peers' messages before committing. We will release the API key, clinic, and personal assistant environments as MAMUBench, comprising 74 scenarios for evaluating multi-user, multi-agent coordination.

Autonomous red team agents increasingly stress-test AI-enabled cyber defenses by planning strategy and executing multistage attacks. Reinforcement learning (RL) and large language models (LLMs) offer complementary mechanisms for the planning and execution such agents require, and prior work has combined them in hybrid hierarchies. Yet a given architecture is typically developed and evaluated within a single environment, leaving open whether an observed advantage reflects a generally stronger decision mechanism or merely alignment with a particular setting. We address this gap with a controlled cross-environment comparison of two homogeneous hierarchical red team architectures: an RL planner with an RL executor (RL+RL) and an LLM planner with an LLM executor (LLM+LLM). We evaluate both against expert autonomous defenders in CybORG CAGE-4 and in Cyberwheel at two network scales, across 18 configurations under one unified disruption metric. We find a pronounced environment-dependent inversion. RL+RL wins the compact, densely rewarded CAGE-4 (78.5% disruption success versus 18.0% for the strongest LLM configuration) and the 100-host Cyberwheel network (81.0% versus 50.5%), while a pretrained cybersecurity LLM agent wins the larger, escalation-gated 1010-host Cyberwheel network (55.0% versus 0.0% for RL). A kill-chain analysis explains the inversion through architecture-specific bottlenecks that aggregate success rates conceal.In the 1010-host Cyberwheel network, RL discovers and compromises hosts but stalls at privilege escalation, whereas in CAGE-4, LLM agents obtain privileged access but rarely convert it into operational impact. These results indicate that conclusions drawn in a single environment may not generalize, and that hybrid planner-executor designs should be motivated by specific failure modes rather than the assumption that one architecture is universally preferable.

Background: Protein language models (PLMs) are increasingly used for sequence generation and property prediction, but their size makes fine-tuning and deployment expensive. The effects of quantization and parameter efficient fine-tuning on performance, representations and generation remain insufficiently characterized. Results: We evaluated 4-bit quantization and low-rank adapter fine-tuning (QLoRA) across ESM-2, ESMC, ProtBERT, ProtT5, Ankh, Ankh3 and Profluent-E1. Across protein prediction tasks, many model-task pairs retained more than 90% of full fine-tuning performance. Peak GPU memory savings approached 90% for the largest models, although performance and efficiency varied by model, dataset and training configuration. QLoRA often preserved early-layer representations while inducing task-specific adaptations in middle and late layers, resembling full fine-tuning with smaller representational changes. Training speed and power effects were more varied. For unconditional generation with ProLLaMA, ProtGPT2, ProGen2, ProteinGLM and ESM3, 4-bit quantization largely preserved predicted structural and sequence-level properties, but token-level analysis revealed model-dependent shifts in autoregressive output distributions. Conclusion: QLoRA and 4-bit quantization reduce PLM computational requirements, particularly GPU memory usage. Our results support QLoRA as a first-pass strategy for memory limited adaptation, reserving full fine-tuning for challenging tasks, unstable architectures or low validation recovery. For generative PLMs, sequence-level and structural metrics should be complemented with distributional analysis, since downstream predictions alone may miss quantization-induced shifts. These approaches can broaden access to large-scale protein modelling while requiring model- and task-specific validation.