Weekly Recap

Diverse LLMs, when strategically routed and collaborating, can collectively achieve significantly fairer outputs than any single model acting alone.

LLMs' apparent behavioral fidelity in sequential tasks can mask fundamentally incorrect generative mechanisms, failing to recover latent structure even with extended context.

Unknown traffic objects are detected with 2.6% higher recall than RT-DETR by fusing semantic and visual prompts symmetrically, enabling real-time open-world detection without NMS.

GPU kernel optimization agents can regress by up to 19% when transferring distilled knowledge across DSLs, revealing that source attempt quality, not just success, dictates transfer efficacy.

Folding multi-branch redundancy slashes speculative decoding costs in diffusion LLMs, unlocking up to 2x faster generation.

LLMs' apparent behavioral fidelity in sequential tasks can mask fundamentally incorrect generative mechanisms, failing to recover latent structure even with extended context.

Unknown traffic objects are detected with 2.6% higher recall than RT-DETR by fusing semantic and visual prompts symmetrically, enabling real-time open-world detection without NMS.

GPU kernel optimization agents can regress by up to 19% when transferring distilled knowledge across DSLs, revealing that source attempt quality, not just success, dictates transfer efficacy.

Folding multi-branch redundancy slashes speculative decoding costs in diffusion LLMs, unlocking up to 2x faster generation.

A single verification prompt is insufficient; learning a reliability score from multiple, complementary perspectives unlocks robust self-verification for VLMs.

The internal journey of an LLM's prediction, not just its destination, reveals crucial information about input difficulty and generator identity.

Diverse LLMs, when strategically routed and collaborating, can collectively achieve significantly fairer outputs than any single model acting alone.

Key contribution not extracted.

Directly learning robot action policies from raw video without explicit action labels unlocks scalable world action models.

A 32B model in Marathi performs like a 10B model in English, revealing how model size and reasoning can bridge language resource gaps in cross-lingual transfer.

Achieve state-of-the-art LLM-guided program evolution for under $2, drastically cutting costs for complex optimization tasks.

Autoregressive video models can now perform explicit, goal-directed reasoning by jointly predicting the final outcome and intermediate steps, achieving state-of-the-art performance with 75% fewer training steps.

Rubric-based RL can be misled by "Vacuous Credit," but MetaRubric's adaptive rubric learning achieves substantial accuracy gains on complex QA tasks.

LLMs can recursively rewrite their own solutions from specialized environments into generalizable training data, boosting performance on complex tasks by over 34% and improving pass@3 by up to 17 percentage points.

LLM agents will pick a worse item from a "preferred" source over a better item from a "dispreferred" source two-thirds of the time, revealing a critical bias that can be reduced by providing more information.

A single set of learned manipulation skills transfers to the real world with 78% success, bypassing the need for task-specific fine-tuning or re-training on real data.

Current omnimodal video embeddings struggle with physical reasoning, yet specific physics can be recovered and leveraged to significantly improve the physical fidelity of generated videos.

A few critical denoising steps in diffusion language models dictate the entire output, and training only these "pivots" dramatically boosts performance.

Most web-browsing LLM agents struggle to find obscure, multi-hop information across diverse modalities and languages, even when explicitly designed for web interaction.

Monolingual RAG performance significantly improves when the LLM's internal reasoning language is aligned with the query and retrieved documents, even if the model is more proficient in another language.

Composing aesthetic preference with explicit rubric rewards unlocks substantial gains in text-to-image generation, outperforming existing open-source models by a significant margin.

State-of-the-art text-to-image models, despite photorealism, still exhibit pervasive world-consistency failures that existing metrics miss.

Generative 3D reconstruction now scales from objects to cityscapes with arbitrary resolution, matching per-scene optimization quality without per-scene tuning.

Adapters forget as much as they learn: a simple post-training normalization unlocks significant gains in both specialization and retained capabilities for LoRA.

RLVR optimization pressure, not supervised fine-tuning, is the theoretical driver of unbounded language drift in LLM chains of thought.

A single tokenizer can now slash token counts for non-English scripts by 20-40% without increasing English tokenization cost.

The non-smooth pointwise behavior of linear optimization solvers smooths to a quadratic form under data distribution averaging, enabling closed-form curvature derivation.

Unlocking latent topological structures with discrete Morse theory boosts graph diffusion model performance, offering theoretical stability guarantees.

Imperfect diffusion models can now achieve exact target distribution sampling at inference time by correcting inherent path and endpoint errors.

Asynchronous LLM training can be made robust to stale rollouts by minimizing a novel group-mass capping bias bound, outperforming existing methods on reasoning tasks.

Multimodal agents can achieve >90% reasoning latency reduction and comparable task performance by selecting from pre-defined reasoning steps instead of generating them freely.

Optimal $\mathcal{O}(\varepsilon^{-4-2/p})$ convergence for stochastic bilevel optimization is achieved with a single-loop first-order method, matching a new lower bound.

LLM-driven heuristic design now learns from noisy system logs to uncover hidden system dynamics, outperforming prior methods in real-world resource assignment.

Robot skills can be dramatically improved post-training by autonomously generating practice tasks, diagnosing failures with privileged simulator states, and iteratively refining symbolic skills and system prompts, achieving near-perfect success on novel manipulation tasks.

Persistent world models can now track objects even when they leave the agent's view, thanks to a novel actor-observer decoupling mechanism.

Real-time 3D Gaussian avatar animation is now achievable at 60fps on mobile by distilling complex neural inference into a linear blendshape approximation, reducing CPU cost by 1000x.

Explicitly disentangling and structuring spatial geometry latents via CR-GEO and routed optimization unlocks significantly more effective 3D reasoning from 2D images, pushing SPAR-Bench and SPBench performance to new heights.

Adversarial distillation achieves state-of-the-art few-step visual generation by learning log-density ratios directly via classification, eliminating the need for auxiliary score fitting.

Standard UAD evaluations for brain MRI are brittle, with a single axis-order mismatch capable of collapsing voxel AUROC from 0.873 to 0.583, while slice-level AUROC remains largely unaffected.

Achieve precise, intuitive 3D control over video generation by moving rigid object meshes, eliminating depth ambiguity and viewpoint inconsistencies.

Robot agents achieve 14% higher success rates and use 65% fewer tokens by composing primitives and performing local retries within an interactive code execution framework.

Even the largest open-weight LLMs fail to generate accurate cybersecurity CLI commands over 58% of the time, but fine-tuning with verifiable rewards on KaliBench closes the gap with massive MoE models.

LLM workflows can be optimized without human labels by treating agent assignment and task decomposition as a bidding process, slashing optimization costs.

State costs in Schrödinger bridges on graphs can be analytically incorporated via a Feynman-Kac tilt, enabling exact, non-learned solutions with linear memory growth even for massive networks.

Teacher averaging in MOPD, not just raw gradients, dictates student learning, with Adam's momentum and BF16 precision masking significant parameter shifts.

Dynamically recomputing embeddings at each denoising step in diffusion transformers halves training compute while achieving state-of-the-art FID scores.

Enterprise data agents struggle to perform complex reasoning and consequential actions across massive, multi-table datasets, with even top models failing over 65% of realistic tasks.

LLMs exhibit substantial generalization instability across multiple behavioral axes, suggesting current accuracy-focused benchmarks mask critical performance disparities.

Humanoid robots can learn entirely new complex manipulation skills at test-time by evolving reward programs, unlocking latent capabilities of pre-trained controllers.

Post-training LLMs for agentic tasks can drastically reduce solution coverage, a "Sharpening Tax" that a novel Bayesian sampler can mitigate.

A probabilistic deep learning emulator for a stochastic climate model not only achieves high fidelity but also learns a latent space that unsupervisedly disentangles metastable regimes and transition dynamics.

Molecular generation models can now achieve 100% validity and state-of-the-art distributional alignment by serializing higher-order topology into context-free grammar rules.

Language model "self-repair" is a predictable consequence of pre-existing counterweight mechanisms, not emergent compensation, governed by an affine law.

Topology-preserving 3D generation now achieves 98% fewer tokens and 58.5% faster inference by encoding shapes as overlapping slice latents instead of expensive voxels.

LLMs' impressive math performance hides a critical "Discovery" bottleneck, but targeting this primitive via self-distillation unlocks significant latent reasoning capacity.

Current interpretability objectives can misrank circuits, favoring less faithful ones that mimic model behavior imperfectly, even when identical circuits are available.

Non-convex sampling from matrix spaces is now possible with provable exponential convergence using momentum preconditioning that spectrally tames gradients.

Robust optimal policies and values for RMDPs are now computable in strongly polynomial time via a single LP, unifying and extending prior results across diverse uncertainty models.

Multimodal sampling with Wasserstein gradient flows and forward diffusion is fundamentally limited by slow mixing, requiring exponential time to bridge well-separated modes.

A novel meta-network framework for data selection achieves unprecedented transferability across diverse LLM training tasks by overcoming the limitations of existing MTS objectives.

A single controller can now robustly learn stochastic transport across a range of initial distribution shifts, outperforming standard SB and noise augmentation on challenging tasks.

A novel hybrid deep learning approach achieves robust few-shot malware detection by leveraging autoencoder-derived features for rapid meta-learning adaptation.

A single CatBoost model detects wildfires 26 times earlier than GOES FDC and VIIRS, and achieves significantly higher F1 scores, even at night.

Unlearning PII from VLMs is now feasible without sacrificing retained knowledge by directly manipulating attention-value representations.

Adversarial domain adaptation that explicitly conditions on predicted weather unlocks robust street-view perception even with non-street-view training data.

Malware analysis just got a new tool: S4 models unlock long-range dependency detection in execution flows, outperforming traditional deep learning for multi-class classification.

Abstract future prediction in latent space significantly boosts LLMs' chemical reasoning and molecular outcome prediction capabilities.

A single token can bypass most safety filters.

Standard GNNs and Transformers can now capture complex multi-way interactions by simply adding higher-order positional encodings derived from clique complexes, boosting performance on tasks requiring rich relational understanding.

Judgment models can solve complex problems when the answer is directly evaluable from the input, but fail catastrophically when simulation is required, even if they possess the underlying knowledge.

A simple "safe response centroid" embedding fails to reliably distinguish safe from unsafe LLM outputs, often performing worse than chance and requiring an explicit unsafe reference to orient the safety direction.

A handful of verified poisoned examples, not just clean ones, dramatically boosts data poisoning defense against clean-label attacks.

A novel Four-Dimensional Architecture Selection Framework correctly recommends 70% of deployed autonomous driving architectures, offering critical guidance for system design.

Learned attention weights actively harm in-context learning performance in Chronos-2, while simple uniform pooling provides consistent benefits.

LLMs can be tricked into leaking sensitive data to attackers simply by being asked to fetch information from seemingly legitimate websites.

Relational learning with tabular foundation models can now bypass feature explosion and interaction blindness by propagating information directly through the schema graph.

Discover the "why" behind ML benchmark gains with MECHVAR, a variance-guided approach that achieves near-optimal mechanism discrimination 100x faster than EIG.

Adversarial training on Fourier phase perturbations unlocks over 20% domain generalization gains in medical imaging by decoupling spatial structure from domain-specific textures.

Mutation-enhanced skill evolution unlocks significant improvements in Lean theorem proving by intelligently sampling mathematical concepts when proof generation stalls.

Discrete neural samplers achieve fixed-point convergence and drastically reduce mode collapse by leveraging Gibbs sampling within a masked diffusion framework.

Expert pairwise rankings, not just absolute scores, can efficiently guide scientific discovery, even when experts have varying fidelity and cost.

Diffusion models can achieve fairness and diversity simultaneously without sensitive attribute labels by using a frozen model adapter and a vision-language embedding space.

Model-free Q-learning now guarantees optimal reachability policies in a broad class of MDPs, slashing sample complexity and memory needs.

Adaptive model search can now achieve statistically controlled adoption without sacrificing evidence efficiency, even with path-dependent challenger generation.

Geographic context unlocks precise PCS anomaly detection, revealing infrastructure health issues previously missed by purely visual methods.

Generating high-quality, diverse datasets from scratch is now possible, outperforming frontier models by 17% in quality and 19% in diversity.

Verifier-guided RL unlocks a 34% accuracy boost in mathematical reasoning by directly penalizing incorrect intermediate steps, not just final answers.

Kernel learners outperform feature learners for nonlinear in-context learning when task diversity is high and context length is short, but the opposite holds for low diversity and long contexts.

Per-user normalization of wrist EDA data can inflate affect recognition performance by up to 4x, masking the true benefit of pretraining and making published results incomparable.

Artists can now directly sculpt 3D object motion relative to a moving camera, overcoming the inherent ambiguity of 2D controls in video generation.

LLMs can learn effective dynamic embeddings for massive combinatorial optimization problems with a shallow policy by selectively propagating memory across rollout steps, bypassing the need for expensive full recomputation or external labels.

LLM agents can now build lasting, reusable "competence" for specific knowledge sources, moving beyond simple retrieval to deep, adaptive understanding.

Two fixed attention blocks suffice for universal interpolation in deep residual networks, even with causal masking.

Most video generation models fail to render text accurately, with even top performers exhibiting a 25% word error rate on a new benchmark designed for this critical oversight.

Quantum devices can exponentially increase their observable complexity over time with finite memory, outperforming classical systems by leveraging phase rotations to achieve perfect discrimination.

Adaptive tensor product computation slashes MLIP runtime by $2.4\times$ and memory by $2.6\times$ without sacrificing accuracy, revealing computational depth's link to chemical complexity.

Current egocentric video generation models fail to execute multi-step manipulation tasks realistically, often skipping crucial actions and exhibiting poor physical dynamics.

Harnessing earlier recurrent states as implicit weak supervision, a training-free contrastive decoding method boosts Looped Transformer performance by up to 11.45% and enables 22.5%-48.2% FLOP reduction.

Transformer and RNN tractography models achieve state-of-the-art performance by introducing a novel generation-validation phase for streamline-level supervision, overcoming the limitations of local loss functions.

LLM training CI is now 77.5% faster and 63.9% more GPU-efficient by intelligently pruning and prioritizing tests based on runtime evidence.