Paradigm Shift

Introduces the concept of a 'trainable' knowledge base for RAG that improves performance by distilling and writing back compact knowledge units.

Uses cycle-consistency as a label-free reward signal for reinforcement learning to resolve contradictions in multimodal reasoning.

Logical reasoning in LLMs is causally linked to 'algebraic divergence' in the residual stream, and failure to achieve this geometry explains sycophancy.

Environment Maps nearly double the success rate of long-horizon agents by replacing session-bound context with a persistent, structured graph representation.

A statistical physics framework that predicts the fundamental limits of agentic self-improvement and nested LLM architectures.

Bio-inspired visual servoing that achieves low-latency robotic control by processing event-stream flux directly, bypassing traditional state estimation.

A massive empirical study of 177,000 tools reveals a rapid shift in the AI agent ecosystem from 'perception' to 'action' (27% to 65% usage).

A simple perturbation method reveals that representations are not just activation patterns, but conduits that determine how learning 'infects' similar examples.

LLMs can solve planning problems with state spaces as large as 10^165 by acting as program generators rather than direct planners.

LLM-generated summaries can produce patient embeddings that are more 'portable' and robust to hospital distribution shifts than specialized clinical models.

Formalizes 'likelihood hacking,' a failure mode where RL-trained models learn to generate unnormalized probabilistic programs to artificially inflate rewards.

A model-agnostic framework to boost time-series forecasting by aligning internal representations with those of pretrained foundation models.

Unifies input and predicted meshes under a shared topological framework to enable high-fidelity 3D reconstruction with sharp features.

Quantifies an emergent 'self' in robots as an invariant subnetwork that persists across continual learning of variable tasks.

Moves automated research from stateless linear pipelines to a persistent Research World Model that maintains a self-correcting knowledge graph of gaps and methods.

Introduces a 'sorry-driven' formal decomposition that allows LLM agents to solve complex proofs by isolating and independently verifying subgoals.

Enforces hard incompressibility constraints in neural operators using spectral Leray projection, ensuring physically admissible fluid simulations.

LensWalk introduces a 'reason-plan-observe' loop that allows agents to dynamically control the temporal sampling and density of the videos they analyze.

The Free-Market Algorithm (FMA) is a zero-parameter metaheuristic that discovers complex pathways in chemistry and economics through emergent supply-and-demand dynamics.

MARCH eliminates 'LLM-as-a-judge' confirmation bias by using information asymmetry to force verification agents to check claims without seeing the original response.

Mechanistic interpretability reveals that LLMs possess 'affect reception' circuits that detect emotional content even when explicit keywords are removed.

Gradient boosting exhibits a 'first-mover bias' where correlated features selected early in the tree sequence gain an artificial, self-reinforcing importance in SHAP rankings.

Establishes a formal mathematical equivalence between Classifier-Free Guidance (CFG) and alignment-based objectives, allowing for CFG-like quality without inference-time overhead.

Shifts symbolic regression from discrete genetic search to a continuous, embedding-driven optimization paradigm.

Replaces standard autoregressive document OCR with a parallel diffusion-based denoising framework.

Demonstrates that Hebbian plasticity can induce emergent attractor dynamics in robot controllers, enabling rapid adaptation without backpropagation.

Instead of using top-activating examples, this method steers Sparse Autoencoder (SAE) features in Vision-Language Models to let the model describe its own internal visual features.

DeIllusionLLM introduces task-level autoregressive reasoning to prevent LLMs from hallucinating answers to ill-posed or faulty scientific questions.

Inter-Layer Structural Encoders (ILSE) use Cayley graphs to aggregate features from all internal LLM layers, improving accuracy by up to 44% over final-layer-only predictions.

Introduces Dual Q-DM, the first non-adversarial imitation learning method theoretically guaranteed to eliminate compounding errors.

Moving beyond coarse reward signals, this paper introduces token-level policy optimization for multimodal reasoning.

This paper moves LLMs from point predictions to set-valued predictions with rigorous statistical coverage guarantees.

Connects stochastic optimal control to the Schrödinger equation, enabling analytic solutions for long-horizon problems that previously scaled exponentially with difficulty.

Enables 3D medical image segmentation pre-training using only mathematical formulas and implicit functions, requiring zero real-world data or expert annotations.

A dual-path architecture that combines speculative speech-to-speech prefixes with cascaded LLM continuations for zero-latency, high-quality dialogue.

A biology-native transformer architecture that mirrors cellular transcription and translation, enabling interpretable predictions across DNA, RNA, and protein.

Introduces a 'geospatial model foundry' that learns unified representations from the weights of existing models rather than raw data.

Enables training of monocular novel-view synthesis models using entirely unpaired, in-the-wild internet images.

Provides a statistically rigorous framework to evaluate model performance and reliability after cherry-picking or selecting models based on the same test data.

Introduces a training strategy where Transformers 'think' in latent space before committing to discrete tokens.

The first foundation model for zero-shot prediction of joint probability distributions in coupled time series.

Formalizes 'Introspection' in LLMs and proves they have privileged access to their own policy logic beyond mere self-simulation.

Reason-to-Transmit introduces deliberative communication for multi-agent systems, where agents reason about *why* a message benefits the receiver rather than just broadcasting features.

This paper demonstrates that Model Context Protocol (MCP) can outperform traditional RAG for quantitative financial Q&A by interacting directly with structured data APIs.

Leum-VL-8B introduces a structural 'grammar' for video parsing by decomposing content into six film-production-style dimensions like camera language and editing.

This paper shows that pretrained monocular models can perform multi-view human mesh recovery without camera calibration or multi-view training data.

Latent representations of reasoning survive cross-architecture translation, allowing student models to inherit teacher capabilities without training.

Coding agents navigating a file system outperform SOTA long-context LLMs and RAG systems on massive datasets.

Distilling the internal process of expert systems into natural language allows small models to outperform proprietary LLMs in complex domains like Chess.

ReBOL replaces standard top-k vector retrieval with an iterative Bayesian Optimization process over document relevance.