Wolf Digest — Saturday, May 9, 2026

Modern retrieval systems, whether lexical or semantic, expose a corpus through a fixed similarity interface that compresses access into a single top-k retrieval step before reasoning. This abstraction is efficient, but for agentic search, it becomes a bottleneck: exact lexical constraints, sparse clue conjunctions, local context checks, and multi-step hypothesis refinement are difficult to implement by calling a conventional off-the-shelf retriever, and evidence filtered out early cannot be recovered by stronger downstream reasoning.

Large language models have achieved remarkable success under the autoregressive paradigm, yet high-quality text generation need not be tied to a fixed left-to-right order. Existing alternatives still struggle to jointly achieve generation efficiency, scalable representation learning, and effective global semantic modeling. We propose Cola DLM, a hierarchical latent diffusion language model that frames text generation through hierarchical information decomposition.

Audio-Visual Intelligence (AVI) has emerged as a central frontier in artificial intelligence, bridging auditory and visual modalities to enable machines that can perceive, generate, and interact in the multimodal real world. In the era of large foundation models, joint modeling of audio and vision has become increasingly crucial, i.e., not only for understanding but also for controllable generation and reasoning across dynamic, temporally grounded signals.

Reinforcement learning with verifiable rewards, particularly Group Relative Policy Optimization (GRPO), has significantly advanced the reasoning capabilities of Large Language Models (LLMs). However, in complex tasks, GRPO frequently suffers from the ``zero-advantage problem'': when all sampled rollouts for a query fail, the relative advantage collapses to zero. Consequently, the model loses effective training signals for these questions, wasting the training data and computational budget. While simply increasing the sampling budget for these questions is a common remedy, the static sampling policy inherently constrains reasoning exploration, limiting the success rate.

Step distillation has become a leading technique for accelerating diffusion models, among which Distribution Matching Distillation (DMD) and Consistency Distillation are two representative paradigms. While consistency methods enforce self-consistency along the full PF-ODE trajectory to steer it toward the clean data manifold, vanilla DMD relies on sparse supervision at a few predefined discrete timesteps.

A growing body of work in cognitive science suggests that reportable conscious access is associated with global ignition over distributed memory systems, while such activation is only partially accessible as individuals cannot directly access or enumerate all activated contents. This tension suggests a plausible mechanism that cognition may rely on a compact representation that approximates the global influence of activation on downstream processing. Inspired by this idea, we introduce the concept of Mindscape Activation Signature (MiA-Signature), a compressed representation of the global activation pattern induced by a query.

We present our winning system for Task~B (generation with reference passages) in SemEval-2026 Task~8: MTRAGEval. Our method is a heterogeneous ensemble of seven LLMs with two prompting variants, where a GPT-4o-mini judge selects the best candidate per instance. We ranked 1st out of 26 teams, achieving a conditioned harmonic mean of 0.7827 and outperforming the strongest baseline (gpt-oss-120b, 0.6390). Ablations show that diversity in model families, scales, and prompting strategies is essential, with the ensemble consistently beating any single model.

Reinforcement learning fine-tuning has become the dominant approach for aligning diffusion models with human preferences. However, assessing images is intrinsically a multi-dimensional task, and multiple evaluation criteria need to be optimized simultaneously. Existing practice deal with multiple rewards by training one specialist model per reward, optimizing a weighted-sum reward R(x)=sum_k w_k R_k(x), or sequentially fine-tuning with a hand-crafted stage schedule. These approaches either fail to produce a unified model that can be jointly trained on all rewards or necessitates heavy manually tuned sequential training.

World Action Models (WAMs) have recently emerged as a promising paradigm for robotic manipulation by jointly predicting future visual observations and future actions. However, current WAMs typically execute a fixed number of predicted actions after each model inference, leaving the robot blind to whether the imagined future remains consistent with the actual physical rollout. In this work, we formulate adaptive WAM execution as a future-reality verification problem: the robot should execute longer when the WAM-predicted future remains reliable, and replan earlier when reality deviates from imagination.

LLM-based agents are increasingly deployed to handle streaming tasks, yet they often remain one-off problem solvers that fail to learn from past interactions. Reusable skills distilled from experience provide a natural substrate for self-evolution, where high-quality skill curation serves as the key bottleneck. Existing approaches either rely on manual skill curation, prescribe heuristic skill operations, or train for short-horizon skill operations. However, they still struggle to learn complex long-term curation policies from indirect and delayed feedback.

Large language models (LLMs) are increasingly used as interactive agents, but optimizing them for long-horizon decision making remains difficult because current methods are largely purely reactive, which weakens both exploration and credit assignment over extended trajectories. In this work, we present Strategic Trajectory Abstraction (StraTA), a simple framework that introduces an explicit trajectory-level strategy into agentic reinforcement learning (RL).

Reinforcement learning with verifiable rewards (RLVR) has become a central paradigm for improving reasoning and code generation in large language models, and GRPO-style training is widely adopted for its simplicity and effectiveness. However, an important design choice remains underexplored: how token-level policy gradient terms are aggregated within each sampled group. Standard GRPO uses sequence aggregation, while recent work has advocated token aggregation as a better alternative.

Reinforcement learning (RL) has been applied to improve large language model (LLM) reasoning, yet the systematic study of how training scales with task difficulty has been hampered by the lack of controlled, scalable environments. We introduce ScaleLogic, a synthetic logical reasoning framework that offers independent control over two axes of difficulty: the depth of the required proof planning (i.e., the horizon) and the expressiveness of the underlying logic.

This paper introduces EnergyFlow, a framework that unifies generative action modeling with inverse reinforcement learning by parameterizing a scalar energy function whose gradient is the denoising field. We establish that under maximum-entropy optimality, the score function learned via denoising score matching recovers the gradient of the expert's soft Q-function, enabling reward extraction without adversarial training. Formally, we prove that constraining the learned field to be conservative reduces hypothesis complexity and tightens out-of-distribution generalization bounds.

We study auto research as a closed empirical loop driven by external measurement. Each submitted trial carries a hypothesis, an executable code edit, an evaluator-owned outcome, and feedback that shapes the next proposal. The output is not a generated paper or a single model checkpoint, but an auditable trajectory of proposals, code diffs, experiments, scores, and failure labels. We instantiate this loop with specialist agents that partition recipe surfaces and share measured lineage across trials.

Reinforcement learning for agentic large language models (LLMs) typically relies on a sparse, trajectory-level outcome reward, making it difficult to evaluate the contribution of individual tool-calls within multi-turn interactions. Existing approaches to such process credit assignment either depend on separate external process reward models that introduce additional consumption, or tree-based structural rollout that merely redistributes the outcome signal while constraining trajectory diversity. A promising alternative leverages the per-turn change in the policy's predicted probability of the ground-truth, termed Information Gain (IG), as an intrinsic process signal without an external evaluator.

Modern Mixture-of-Experts (MoE) architectures allocate expert capacity through a rigid per-layer rule: each transformer layer owns a separate expert set. This convention couples depth scaling with linear expert-parameter growth and assumes that every layer needs isolated expert capacity. However, recent analyses and our routing probe challenge this allocation rule: replacing a deeper layer's learned top-k router with uniform random routing drops downstream accuracy by only 1.0-1.6 points across multiple production MoE models.

Foundation models have established unified representations for natural language processing, yet this paradigm remains largely unexplored for tabular data. Existing methods face fundamental limitations: LLM-based approaches lack retrieval-compatible vector outputs, whereas text embedding models often fail to capture tabular structure and numerical semantics. To bridge this gap, we first introduce the Tabular Embedding Benchmark (TabBench), a comprehensive suite designed to evaluate the tabular understanding capability of embedding models. We then propose TabEmbed, the first generalist embedding model that unifies tabular classification and retrieval within a shared embedding space.

Geospatial reasoning requires models to resolve complex spatial semantics and user intent into precise target locations for Earth observation. Recent progress has liberated the reasoning path from manual curation, allowing models to generate their own inference chains. Yet a final dependency remains: they are still supervised by human-annotated ground-truth coordinates. This leaves the reasoning process autonomous, but not its spatial endpoint, and prevents true self-evolution on abundant unlabeled remote sensing data. To break this bottleneck, we introduce RemoteZero, a box-supervision-free framework for geospatial reasoning.

Large language models (LLMs) are routinely prompted to take on social roles ranging from individuals to institutions, yet it remains unclear whether their internal representations encode the granularity of such roles, from micro-level individual experience to macro-level organizational, institutional, or national reasoning. We show that they do. We define a contrast-based Granularity Axis as the difference between mean macro- and micro-role hidden states.

LLM-based Triton kernel generation has attracted significant interest, yet a fundamental empirical question remains unanswered: where does this capability break down, and why? We present KernelBench-X, a benchmark designed to answer this question through category-aware evaluation of correctness and hardware efficiency across 176 tasks in 15 categories. Our systematic comparison of five representative methods yields three main findings. First, task structure determines correctness more than method design.

We investigate the scaling properties of implicit deductive reasoning over Horn clauses in depth-bounded Transformers. By systematically decorrelating provability from spurious features and enforcing algorithmic alignment, we find that in sufficiently deep models with a bidirectional prefix mask, implicit reasoning approaches explicit CoT performance across graph topologies and problem widths, though CoT remains necessary for depth extrapolation.

Despite the growing popularity of Multimodal Domain Generalization (MMDG) for enhancing model robustness, it remains unclear whether reported performance gains reflect genuine algorithmic progress or are artifacts of inconsistent evaluation protocols. Current research is fragmented, with studies varying significantly across datasets, modality configurations, and experimental settings. Furthermore, existing benchmarks focus predominantly on action recognition, often neglecting critical real-world challenges such as input corruptions, missing modalities, and model trustworthiness. This lack of standardization obscures a reliable assessment of the field's advancement.

Recent advances in generative video models are increasingly driven by post-training and test-time scaling, both of which critically depend on the quality of video reward models (RMs). An ideal reward model should predict accurate rewards that align with human preferences across diverse scenarios. However, existing paradigms face a fundamental dilemma: Discriminative RMs regress rewards directly on features extracted by multimodal large language models (MLLMs) without explicit reasoning, making them prone to shortcut learning and heavily reliant on massive data scaling for generalization.

Many deployments must compare candidate language models for safety before a labeled benchmark exists for the relevant language, sector, or regulatory regime. We formalize this setting as benchmarkless comparative safety scoring and specify the contract under which a scenario-based audit can be interpreted as deployment evidence. Scores are valid only under a fixed scenario pack, rubric, auditor, judge, sampling configuration, and rerun budget.

Multimodal Large Language Models (MLLMs) hold significant promise for revolutionizing traditional education and reducing teachers' workload. However, accurately interpreting unconstrained STEM student handwritten solutions with intertwined mathematical formulas, diagrams, and textual reasoning poses a significant challenge due to the lack of authentic and domain-specific benchmarks. Additionally, current evaluation paradigms predominantly rely on the outcomes of downstream tasks (e.g., auto-grading), which often probe only a subset of the recognized content, thereby failing to capture the MLLMs' understanding of complex handwritten logic as a whole.

David Reich is back. He and collaborator Ali Akbari just published a paper that overturns a long-standing consensus about human evolution — that natural selection has been dormant in our species since the agricultural revolution. By scaling ancient DNA sequencing and developing a new statistical method, they found that selection has actually sped up. Selection went especially bonkers during the Bronze Age (around 3,000 years ago). That's when gene frequencies for everything from immune function to body fat to intelligence were most in flux.

We introduce ReflectDrive-2, a masked discrete diffusion planner with separate action expert for autonomous driving that represents plans as discrete trajectory tokens and generates them through parallel masked decoding. This discrete token space enables in-place trajectory revision: AutoEdit rewrites selected tokens using the same model, without requiring an auxiliary refinement network. To train this capability, we use a two-stage procedure. First, we construct structure-aware perturbations of expert trajectories along longitudinal progress and lateral heading directions and supervise the model to recover the original expert trajectory.

We introduce the AI co-mathematician, a workbench for mathematicians to interactively leverage AI agents to pursue open-ended research. The AI co-mathematician is optimized to provide holistic support for the exploratory and iterative reality of mathematical workflows, including ideation, literature search, computational exploration, theorem proving and theory building. By providing an asynchronous, stateful workspace that manages uncertainty, refines user intent, tracks failed hypotheses, and outputs native mathematical artifacts, the system mirrors human collaborative workflows.

High-resolution image-to-video (I2V) generation aims to synthesize realistic temporal dynamics while preserving fine-grained appearance details of the input image. At 2K resolution, it becomes extremely challenging, and existing solutions suffer from various weaknesses: 1) end-to-end models are often prohibitively expensive in memory and latency; 2) cascading low-resolution generation with a generic video super-resolution tends to hallucinate details and drift from input-specific local structures, since the super-resolution stage is not explicitly conditioned on the input image. To this end, we propose SwiftI2V, an efficient framework tailored for high-resolution I2V.

Training compute is increasingly outpacing the availability of high-quality data. This shifts the central challenge from optimal compute allocation to extracting maximum value from limited data. The widely adopted Chinchilla scaling law assumes every training token is unique. This limits its ability to guide pretraining decisions in data-constrained regimes. We model the excess loss under repetition with a simple additive overfitting penalty and find that it accurately describes model behavior. Our scaling law yields qualitatively new compute-optimal allocation advice.

We revisit a universally accepted but under-examined design choice in every modern LLM: a token index is looked up once at the input embedding layer and then permanently discarded. This single-injection assumption induces two structural failures: (i) the Rare Token Problem, where a Zipf-type distribution of vocabulary causes rare-token embeddings are chronically under-trained due to receiving a fraction of the cumulative gradient signal compared to common tokens; and (ii) the Contextual Collapse Problem, where limited parameters models map distributionally similar tokens to indistinguishable hidden states.

Symbolic music datasets with matched scores and performances are essential for many music information retrieval (MIR) tasks. Yet, existing resources often cover a narrow range of composers, lack performance variety, omit note-level alignments, or use inconsistent naming formats. This work presents PianoCoRe, a large-scale piano MIDI dataset that unifies and refines major open-source piano corpora. The dataset contains 250,046 performances of 5,625 pieces written by 483 composers, totaling 21,763 h of performed music.

We address the challenge of knowledge composition in Vision-Language Models (VLMs), where accumulating expertise across multiple domains or tasks typically leads to catastrophic forgetting. We introduce GeoStack (Geometric Stacking), a modular framework that allows independently trained domain experts to be composed into a unified model. By imposing geometric and structural constraints on the adapter manifold, GeoStack ensures the foundational knowledge of the base model is preserved. Furthermore, we mathematically demonstrate a weight-folding property that achieves constant-time inference complexity (O(1)), regardless of the number of integrated experts.

High-performance computing (HPC) is increasingly important for scalable quantum chemistry workflows that couple classical generative models, quantum circuit simulation, and selected configuration interaction postprocessing. We present the generative quantum-inspired Kolmogorov-Arnold eigensolver (GQKAE), a parameter-efficient extension of the generative quantum eigensolver (GQE) for quantum chemistry. GQKAE replaces the parameter-heavy feed-forward network components in GPT-style generative eigensolvers with hybrid quantum-inspired Kolmogorov-Arnold network modules, forming a compact HQKANsformer backbone.

Despite the success of large language models (LLMs) on general-purpose tasks, their performance in highly specialized domains such as biomedicine remains unsatisfactory. A key limitation is the inability of LLMs to effectively leverage biomedical tools, which clinical experts and biomedical researchers rely on extensively in daily workflows. While recent general-domain tool-calling datasets have substantially improved the capabilities of LLM agents, existing efforts in the biomedical domain largely rely on in-context learning and restrict models to a small set of tools.

In recent years, open-source efforts like Senorita-2M have propelled video editing toward natural language instruction. However, current publicly available datasets predominantly focus on local editing or style transfer, which largely preserve the original scene structure and are easier to scale. In contrast, Background Replacement, a task central to creative applications such as film production and advertising, requires synthesizing entirely new, temporally consistent scenes while maintaining accurate foreground-background interactions, making large-scale data generation significantly more challenging. Consequently, this complex task remains largely underexplored due to a scarcity of high-quality training data.

The Pentagon's inspector general recently met with the Justice Department's first-ever assistant attorney general for the national fraud enforcement division to explore opportunities for their teams to cooperate more closely on efforts to confront increasingly complex scams impacting the Defense Department and military.

MIT Technology Review Explains : Let our writers untangle the complex, messy world of technology to help you understand what's coming next. You can read more from the series here . Eight passengers aboard a Dutch-flagged cruise ship have contracted a type of hantavirus, a rare virus transmitted by rats. Three of them have died. As the ship prepares to dock in the Canary Islands, plans are being finalized to let the remaining passengers and crew disembark safely. The virus in question appears to have a high fatality rate.

Liang Wenfeng, billionaire founder and CEO of DeepSeek, is planning to write the biggest check for the startup's first-ever funding round, which the company now hopes will raise up to 50 billion yuan ($7.35 billion), according to two people with direct knowledge of the talks. That would make it the biggest funding round by a Chinese AI company ever. Meanwhile, the funding round has prompted the Chinese AI lab to expedite its plans to generate revenue and become commercially viable.

The Marine Corps is overhauling how it trains its reconnaissance troops as ubiquitous surveillance tools redefine modern conflict and service officials wrestle with difficult questions over sensor and robotic employment for new training. Two new courses replaced the Basic Reconnaissance Course, a 12-week program known for its grueling nature in a move service officials said was meant to modernize training, reduce wait times for advanced schools and strengthen baseline infantry skills to meet fleet demands.

The Pentagon's counter-drone task force announced a new pilot program this week aimed at fielding directed energy systems for UAS defense to five military installations across the country over the next six months. Joint Interagency Task Force 401, an Army-led entity charged with boosting the military's counter-drone efforts, said the initiative is intended to protect infrastructure, military installations and domestic missions against unmanned aerial systems.

CyberSecQwen-4B: Why Defensive Cyber Needs Small, Specialized, Locally-Runnable Models

At a glance We construct geographically grounded, electrically coherent power grid models entirely from publicly available data and release a dataset spanning 48 U.S. states and multi-state interconnections. The models support AC optimal power flow (AC‑OPF) analysis, enabling physics-based study of congestion, capacity, and demand siting without restricted data. We demonstrate applications including transmission expansion potential, targeted line upgrades, and placement of large datacenter loads. Microsoft Research is excited to release an open dataset of approximate transmission topology of the U.S. power grid derived from publicly available data.

In the second week of the landmark trial between Elon Musk and OpenAI, Musk's motivations for bringing the suit were under scrutiny. Last week, Musk took the stand, alleging that OpenAI CEO Sam Altman and president Greg Brockman had deceived him into donating $38 million to the company. He claimed that they'd promised to maintain it as a nonprofit dedicated to developing AI for the benefit of humanity, only to later accept billions of dollars of investment from Microsoft and restructure the company to operate a for-profit subsidiary.

This is today's edition of The Download , our weekday newsletter that provides a daily dose of what's going on in the world of technology. We've entered the era of AI malaise AI is spreading everywhere, and it is not going away. But what will it do? What effect will it have on our society? Will it make life better, or worse? How will we know?

Technology is changing the way we make babies. The pioneering work of the scientists who invented IVF led to the birth of the first "test tube baby" in 1978. We've come a long, long way since then. This week, I've been working on a piece about the cutting edge of IVF technologies and what's coming next. Think AI and robots and, potentially, gene-edited embryos. My reporting has also made me think about just how much progress has been made in the last five decades.

Dan Levine, a partner at Accel known for early startup investments in Scale AI and Vercel, is stepping back from the venture capital firm and will no longer make new investments for it, people familiar with the matter said. The firm disclosed the change to limited partners in recent weeks. It ...

Anthropic is the unnamed customer behind a $1.8 billion cloud deal that Akamai announced on Thursday, according to a person with direct knowledge of the matter. That deal sent the content delivery network provider's stock soaring 27% on Friday. The seven-year cloud contract—which Akamai ...

Cursor is already starting to make its presence known at SpaceX's AI unit, just weeks after Elon Musk's firm got an option to buy the coding startup for $60 billion. Cursor staffers have been visiting xAI offices to meet with employees and discuss their work, according to two people with direct knowledge of the companies. Fresh xAI exits have followed, including staff cuts last Friday, the people said.

Salesforce's chief communications officer Carolyn Guss is leaving the sales software giant, according to two people familiar with the matter. One of the people said Guss is joining a new company and that Salesforce will be appointing another chief communications officer soon. Among her ...

Apple has reached "a preliminary agreement" with Intel in recent months for manufacturing some of its chips, The Wall Street Journal reported . Earlier, this week Bloomberg reported that Apple was exploring such a partnership with Intel and Samsung. A deal with one of the companies would help ...

The hedge fund TCI cut almost all of its $8 billion stake in Microsoft, the Financial Times reported Friday , due to what CEO Christopher Hohn described in an investor letter as "uncertainty over Microsoft's competitive position in the future" as AI threatens to replace existing software like ...

Shares of iconic glassmaker Corning Inc. have doubled so far this year to their highest price ever, thanks to surging demand for Corning's fiber optic cables that connect servers and AI data centers to one another. A partnership Corning announced with Nvidia this week gave the stock another lift. But Corning's current deluxe valuation, which is now based almost entirely on sky-high expectations for its optics business, might be more fragile than it appears.

❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambda.ai/papers 📝 GPT 5.5 Instant: https://deploymentsafety.openai.com/gpt-5-5-instant/introduction https://openai.com/index/gpt-5-5-instant/ Classifiers paper: https://arxiv.org/pdf/2501.18837 Our Patreon if you wish to support us: https://www.patreon.com/TwoMinutePapers 🙏 We would like to thank our generous Patreon supporters who

(Photo by Maddie Meyer/Getty Images) Welcome back to This Week in Stratechery! As a reminder, each week, every Friday, we're sending out this overview of content in the Stratechery bundle; highlighted links are free for everyone . Additionally, you have complete control over what we send to you. If you don't want to receive This Week in Stratechery emails (there is no podcast), please uncheck the box in your delivery settings . On that note, here were a few of our favorites this week.

The drones hitting Gulf Arab states daily since the United States and Israel launched large-scale military operations against Iran in February are not merely Iranian. They are originally Iranian, yes. But these designs and production processes were improved and refined by Russia through years of battlefield testing against Ukrainian defenses. So, they were returned to Tehran from Moscow. Confronted with a threat that Ukraine has spent four years learning to counter, the United States found itself in unfamiliar territory.

Some found out they didn't qualify for WARN Act protections like two-months notice because the company had classified them as remote workers.

Intel's stock has risen a stunning 490% over the past year, a bet by Wall Street that may be running well ahead of the company's actual turnaround.

Cloudflare announced its first large-scale layoff. CEO Matthew Prince says because of AI efficiency gains, the company doesn't need as many support roles.

Everyone wants a piece of the enterprise AI pie, and this week, we saw a string of companies making their moves. From Anthropic and OpenAI announcing new joint ventures targeting enterprise AI deployment to SAP dropping $1B on German AI startup Prior Labs, it's becoming clear that if you're a startup building enterprise tools, you're likely an acquisition target. On this episode of TechCrunch's Equity podcast, hosts Kirsten Korosec, Anthony

Last day to buy one pass and get a second one at 50% off to TechCrunch Disrupt 2026. Bring a partner, co-founder, or colleague at half off. Register now.