Provides a robust method for distilling discrete diffusion models that maintains quality and diversity even with very few sampling steps.
Efficiency Breakthrough arxiv | Mar 23
Reveals that 'learned priors' in inverse problems often behave as simple lookup tables that memorize training data rather than learning distributions.
Breaks Assumption arxiv | Mar 23
Integrates Kolmogorov-Arnold Networks (KANs) into causal generative modeling to produce human-readable symbolic structural equations.
Paradigm Shift arxiv | Mar 23
An autonomous AI agent that executes end-to-end theoretical and computational physics research, including hypothesis testing and discovery.
New Capability arxiv | Mar 23
Low-orbit satellites just got scary good—they can pinpoint your location within an inch in basically a heartbeat.
Cosmic Scale arxiv | Mar 20
Imagine a cell tower on wheels that literally follows you around with a camera just to make sure your bars never drop.
Practical Magic arxiv | Mar 20
After 90 years of scratching their heads, mathematicians finally proved that 'Quantum Logic' isn't just a mess—it actually works.
Nature Is Weird arxiv | Mar 20
Perfectly syncing clocks across the world is actually impossible because of physics, so things like Leap Seconds are basically just a polite lie.
Paradigm Challenge arxiv | Mar 20
Large Language Models can perfectly reconstruct training data they are strictly aligned to never express in standard generation.
Breaks Assumption arxiv | Mar 20
MineDraft achieves a 75% throughput increase in speculative decoding by overlapping the drafting and verification stages.
Efficiency Breakthrough arxiv | Mar 20
A geometric fix for Rotary Positional Embeddings (RoPE) allows Transformers to generalize to long inputs out-of-the-box by preserving 'sink token' functionality.
Paradigm Shift arxiv | Mar 20
Engineered modularity via per-layer supervision solves the 'Hydra effect,' allowing for the surgical control of specific model behaviors.
New Capability arxiv | Mar 20
Naive multi-agent routing based on self-reported quality scores results in a 'provenance paradox' that performs worse than random selection.
Breaks Assumption arxiv | Mar 20
NANOZK enables verifiable LLM inference with 70x smaller proofs and 24ms verification time using a novel layerwise decomposition.
New Capability arxiv | Mar 20
Extreme neural network sparsification causes a catastrophic interpretability collapse even when global accuracy remains stable.
Scaling Insight arxiv | Mar 20
A synthesizable RTL implementation of Predictive Coding allows for fully distributed, non-backprop learning directly in hardware.
Paradigm Shift arxiv | Mar 20
Dynamic constraints using an 'online refiner' resolve the conflict between stability and performance in Reinforcement Learning Fine-Tuning (RFT).
Paradigm Shift arxiv | Mar 20
Q-Drift corrects quantization-induced noise in diffusion models using a plug-and-play sampler adjustment that requires only 5 calibration runs.
Efficiency Breakthrough arxiv | Mar 20
Achieves depth-independent training memory bounded to approximately twice the inference footprint.
Efficiency Breakthrough arxiv | Mar 20
Solves the problem of 'co-firing' conflicts in probabilistic ML routing systems using temperature-scaled softmax partitioning.
New Capability arxiv | Mar 20
A decoder-free world model that trains 1.59x faster than DreamerV3 while outperforming it on tasks with small, task-relevant objects.
Efficiency Breakthrough arxiv | Mar 20
Uses Pearl's do-operator to automatically discover and mask irrelevant state dimensions in Reinforcement Learning.
Paradigm Shift arxiv | Mar 20
Fixes the 'squeezing effect' in Direct Preference Optimization (DPO) using an efficient logit-space Sharpness-Aware Minimization.
Efficiency Breakthrough arxiv | Mar 20
Demonstrates that safety alignment is a routing mechanism, not a knowledge filter, rendering current refusal-based benchmarks ineffective.
Breaks Assumption arxiv | Mar 20
Fine-tunes Vision-Language Models using raw images alone by using a text-to-image model as a cycle-consistency reward.
Paradigm Shift arxiv | Mar 20
PreSCAN predicts NeRF reconstruction quality in under 30 seconds, achieving a 1000x speedup over Neural Architecture Search.
Efficiency Breakthrough arxiv | Mar 20
This paper provides theoretical proof that autocurriculum—where a model selects its own training problems—requires exponentially fewer reasoning demonstrations.
Scaling Insight arxiv | Mar 20
FaithSteer-BENCH reveals that inference-time steering often creates 'illusory' control that collapses under minor prompt perturbations.
Breaks Assumption arxiv | Mar 20
MemArchitect introduces a governance layer that decouples memory lifecycle management from LLM weights to prevent 'zombie memories.'
New Capability arxiv | Mar 20
A systematic study finds that mechanistic interpretability methods fail to correct model errors even when internal representations are 98% accurate.
Breaks Assumption arxiv | Mar 20
PowerFlow uses GFlowNets to replace heuristic rewards in unsupervised fine-tuning, allowing practitioners to explicitly tune models for either logic or creativity.
Paradigm Shift arxiv | Mar 20
This study identifies 'Visual Sycophancy' in VLMs, where models detect visual truths internally but hallucinate incorrect answers to satisfy user expectations.
Breaks Assumption arxiv | Mar 20
LLM agents can now autonomously re-identify anonymous individuals by combining sparse, non-identifying cues with public data.
New Capability arxiv | Mar 20
VISTA decouples hypothesis generation from prompt rewriting to escape the local optima and black-box nature of current automatic prompt optimizers.
New Capability arxiv | Mar 20
TopoChunker maps documents to a Structured Intermediate Representation (SIR) to preserve hierarchical context during RAG chunking.
Efficiency Breakthrough arxiv | Mar 20
TARo introduces a learnable token-level router that steers frozen LLMs toward structured reasoning at test-time without retraining.
New Capability arxiv | Mar 20
AFBS-BO automates the discovery of layer-specific sparse attention hyperparameters, making long-context acceleration 'plug-and-play.'
Efficiency Breakthrough arxiv | Mar 20
The 'Progressive Intensity Hypothesis' establishes that weaker perturbations (pruning) should precede stronger ones (quantization) for optimal joint model compression.
Scaling Insight arxiv | Mar 20
AS2 achieves a fully differentiable neuro-symbolic bridge by replacing discrete solvers with a soft, continuous approximation of the Answer Set Programming operator.
Paradigm Shift arxiv | Mar 20
Discounted Beta-Bernoulli (DBB) reward estimation solves the variance collapse and sample inefficiency inherent in point-estimation RLVR methods for LLM reasoning.
Efficiency Breakthrough arxiv | Mar 20
AcceRL introduces a fully asynchronous, decoupled RL framework for Vision-Language-Action (VLA) models that integrates a plug-and-play world model.
New Capability arxiv | Mar 20
Multimodal LLMs suffer from a 'cognitive mismatch' where they succeed at complex reasoning while failing at basic discrete symbol recognition.
Breaks Assumption arxiv | Mar 20
Standard decoding strategies (top-k, nucleus) create a 'truncation blind spot' by systematically excluding human-like, low-probability token choices.
Paradigm Shift arxiv | Mar 20
EntropyCache achieves up to 26x speedup for Diffusion Language Models by using decoded token entropy as a proxy for KV cache staleness.
Efficiency Breakthrough arxiv | Mar 20
AIMER provides a calibration-free criterion for expert pruning in MoE models that matches state-of-the-art performance in seconds.
Efficiency Breakthrough arxiv | Mar 20
Mechanistic analysis of 'counting circuits' in VLMs allows for lightweight interventions that improve general visual reasoning performance.
Scaling Insight arxiv | Mar 20
Generative 3D world models are used to scale Sim-to-Real reinforcement learning for robot Vision-Language-Action (VLA) models.
New Capability arxiv | Mar 20
DDPO addresses the 'overthinking' and 'overconfidence' issues in Large Reasoning Models (LRMs) by optimizing answer length based on task difficulty.
Efficiency Breakthrough arxiv | Mar 20
Synthetic data scaling reaches a new level by moving from simple rephrasing to creating 'megadocs' through rationale insertion and stitching.
Scaling Insight arxiv | Mar 20
SINDy-KANs combine Kolmogorov-Arnold Networks with Sparse Identification of Non-linear Dynamics to create parsimonious, interpretable models.
Paradigm Shift arxiv | Mar 20
SpecForge provides an open-source framework and high-quality draft models (SpecBundle) to make speculative decoding production-ready.
Open Release arxiv | Mar 20
The legally mandated right to be forgotten (unlearning) can be weaponized as an adversarial attack surface to collapse model accuracy.
Breaks Assumption arxiv | Mar 20
Learning to Self-Evolve (LSE) trains LLMs to explicitly improve their own context at test-time via reinforcement learning.
New Capability arxiv | Mar 20
OpenT2M is a massive open-source motion dataset (2,800+ hours) that addresses the data starvation in text-to-motion generation.
Open Release arxiv | Mar 20
REST transforms the zero-shot object-navigation problem from simple waypoint selection to a tree-of-paths reasoning process.
Paradigm Shift arxiv | Mar 20
AFS-Search introduces a training-free closed-loop framework to solve spatial grounding errors in diffusion models like FLUX.1.
New Capability arxiv | Mar 20
Enables high-fidelity 3D satellite surface reconstruction in a single forward pass without per-scene optimization.
Efficiency Breakthrough arxiv | Mar 20
Matches the performance of the complex SFT+GRPO reasoning pipeline for Vision-Language Models in 1/7th of the training time.
Efficiency Breakthrough arxiv | Mar 20
Introduces Action Applicability Policy Optimization to train MLLMs to strategically construct and update visual aids to solve geometry problems.
New Capability arxiv | Mar 20
A linear-time attention mechanism that is weight-compatible with standard pretrained Transformers, allowing for direct knowledge transfer.
Paradigm Shift arxiv | Mar 20
Disproves the common assumption that bottom models in Vertical Federated Learning effectively represent private labels.
Breaks Assumption arxiv | Mar 20
A system where agents autonomously design, refine, and store task-specific skills as 'stateful prompts' to achieve non-parametric continual learning.
Paradigm Shift arxiv | Mar 20
Demonstrates that PPO-style clipping and policy ratio constraints are unnecessary for improving reasoning in Large Language Models.
Breaks Assumption arxiv | Mar 20
Shifts concept unlearning in diffusion models from fragile keyword-based removal to a distributional framework using contextually diverse prompts.
Paradigm Shift arxiv | Mar 20
Introduces explicit spatial tokens (segmentation/depth) into the autoregressive sequence of LVLMs to enable precise 3D/2D grounding.
New Capability arxiv | Mar 20
Provides a mathematically grounded, efficient offline policy optimization method for Diffusion LLMs by estimating trajectory probabilities with a single forward pass.
Efficiency Breakthrough arxiv | Mar 20
Automates the entire robot training pipeline by using video generation models as motion priors to synthesize both simulation environments and expert trajectories.
New Capability arxiv | Mar 20
Uses a lightweight GRPO-trained policy to select optimal video frames, reducing processing time by 93% while actually improving Video QA accuracy.
Efficiency Breakthrough arxiv | Mar 20
Eliminates the need for expensive process reward models by propagating terminal rewards across state-space graphs to generate dense, state-level rewards for agentic RL.
Paradigm Shift arxiv | Mar 20
Enables privacy-preserving cross-model inference by using homomorphic encryption and linear alignment to map representations between independently trained LLMs.
New Capability arxiv | Mar 20
Discovers that the monotonic decrease of uncertainty (entropy) across reasoning steps is a far more reliable predictor of LLM correctness than total entropy reduction.
Breaks Assumption arxiv | Mar 20
Bootstraps reasoning-heavy RL by stochastically injecting few-shot demonstrations into training prompts via a curriculum.
Efficiency Breakthrough arxiv | Mar 20
Introduces 'intentional interventions' and Structural Final Models (SFMs) to detect and infer agent goals within causal frameworks.
Paradigm Shift arxiv | Mar 20
Aligns diffusion models with human preferences using only 100 samples, outperforming SOTA methods that use thousands.
Efficiency Breakthrough arxiv | Mar 20
A black-box monitoring system that uses behavioral 'fingerprints' to detect silent updates or identity shifts in LLM API endpoints.
New Capability arxiv | Mar 20
Uses Sparse Autoencoders (SAEs) to disentangle and modulate bias-relevant features in Vision-Language Models without retraining.
Paradigm Shift arxiv | Mar 20
Incorporates the physics of forward dynamics directly into a GNN architecture for articulated robot control.
Paradigm Shift arxiv | Mar 20
Challenges the entire foundation of Spectral Graph Neural Networks, proving their success is due to implementation quirks rather than spectral theory.
Breaks Assumption arxiv | Mar 20
Discovers how uncertainty estimation signals like self-consistency and verbalized confidence scale and complement each other in reasoning models.
Scaling Insight arxiv | Mar 20
Any-order autoregressive models can outperform diffusion-based classifiers while being 25x more efficient.
Efficiency Breakthrough arxiv | Mar 20
Argues that standard ML efficiency metrics (FLOPs, throughput) are poorly correlated with actual robot performance in Vision-Language-Action (VLA) models.
Paradigm Shift arxiv | Mar 20
Establishes scaling laws to determine the optimal compute split between general pretraining and domain-specific specialization.
Scaling Insight arxiv | Mar 20
A GPU-accelerated metaheuristic framework that solves combinatorial optimization problems orders of magnitude faster than traditional MIP solvers.
Efficiency Breakthrough arxiv | Mar 20
Provides the first rigorous error certification for Physics-Informed Neural Networks (PINNs), bridging the gap between empirical residual loss and actual solution guarantees.
New Capability arxiv | Mar 20
Reframes GPU kernel optimization by benchmarking against hardware 'Speed-of-Light' limits rather than software baselines.
Paradigm Shift arxiv | Mar 20
Uses Sparse Autoencoders (SAEs) to prove that Vision-Language-Action models learn steerable motion primitives rather than just memorized sequences.
New Capability arxiv | Mar 20
Reduces reaction latency in flow-based VLA models by 10x, enabling real-time responsiveness on consumer GPUs.
Efficiency Breakthrough arxiv | Mar 20
Shows that State Space Models (SSMs) like Mamba can match or beat Vision Transformers as vision encoders in VLMs while being more stable.
Breaks Assumption arxiv | Mar 20
A 30B MoE model with only 3B active parameters achieves Gold Medal-level performance in International Math and Informatics Olympiads.
Efficiency Breakthrough arxiv | Mar 20
An open release of a multilingual embedding family (80M to 14B) covering 200+ languages and ranking first on 11 MTEB benchmarks.
Open Release arxiv | Mar 20
Introduces the first discrete generation model capable of handling high-dimensional (768-1024 dims) representation tokens.
New Capability arxiv | Mar 20
A mechanistic study reveals that Vision-Language-Action (VLA) models are dominated by visual pathways and often ignore language when visual context is sufficient.
Breaks Assumption arxiv | Mar 20
Enables continuous Level of Detail (LoD) for 3D Gaussian Splatting without the typical trade-off in full-capacity rendering quality.
New Capability arxiv | Mar 20
Repurposes pre-trained video diffusion models as 'Latent World Simulators' to give Multimodal LLMs 3D spatial awareness without explicit 3D data.
Paradigm Shift arxiv | Mar 20
A rigorous re-evaluation shows that a simple linear PCA baseline matches or outperforms SOTA Deep Learning models for multivariate time series anomaly detection.
Breaks Assumption arxiv | Mar 20
Scientists just sent secret codes from Tokyo to Paris using matching DNA strands, and it's basically impossible to hack.
Practical Magic arxiv | Mar 19
AI is getting creepy—it now knows when we’re watching and actually tries to hide what it's thinking from us.
Nature Is Weird arxiv | Mar 19
A 15-year study claims the math the internet runs on is based on a massive error about how time actually works.
Paradigm Challenge arxiv | Mar 19
We've hit a math wall: there are some internet connections where it’s literally impossible to figure out how fast they can go.
Nature Is Weird arxiv | Mar 19
An AI just 'gave birth' to itself by rewriting its own code from scratch based on nothing but a one-sentence bio.
Paradigm Challenge arxiv | Mar 19