AI Wiki — Knowledge Base

LangChain 今天连发两记，直接把 agent reliability 往前推了一步：一套统一评估栈，一个不用完整沙箱的代码执行方案。先说评估。长运行、有状态的 agent 怎么测？Harbor + LangSmith 的组合给出一个 runner，能挂载 sandbox…

LangChain 今天连发两记，直接把 agent reliability 往前推了一步：一套统一评估栈，一个不用完整沙箱的代码执行方案。先说评估。长运行、有状态的 agent 怎么测？Harbor + LangSmith 的组合给出一个 runner，能挂载 sandbox、接入 observability，终于不是靠单元测试糊弄了。再看代码执行。Deep Agents 需要跑不受信代码，

agentsharrison-chase--x

30 Jun 2026

All entries

KernelSight-LM: A Kernel-Level LLM Inference Simulator

Keypose Exploration: Efficient Automatic Trajectory Labelling and Cross-Embodiment Policy Transfer

KM-Speaker: Keypoint-Based Style Control for High-Quality Speech-Driven 3D Facial Animation and Dialogue Localization

Know Before You Fetch: Calibrated Retrieval-Budget Allocation for Retrieval-Augmented Generation

Knowing Bias, Doing Better: Mitigating Social Bias in LLMs via Know-Bias Neuron Enhancement

Knowing in Advance When an Evolutionary Outer Loop Will Not Help: A Pre-Registered Cheap-Baseline Screening Rule

KnowsTFM: Knowledge-Informed Fine-Tuning of Small Tabular Foundation Models

Kriging and neural network models for pressure losses across perforated plates

KrishokChat: A Citation-Grounded Dataset and Benchmark for Bengali Agricultural Advisory

KYON: Semi-Modular Wheel-Legged Quadruped With Agile Bimanual Capability

L2D2-GS: Learning to Densify for Feedforward Dynamic Gaussian Scene Reconstruction

Labeling Training Data for Entity Matching Using Large Language Models

LaGen: Towards Autoregressive LiDAR Scene Generation

LAMP: Lean-based Agentic framework with MCP and Proof Repair

LAMP: Long-Horizon Adaptive Manipulation Planning for Multi-Robot Collaboration in Cluttered Space

LangChain 今天连发两记，直接把 agent reliability 往前推了一步：一套统一评估栈，一个不用完整沙箱的代码执行方案。 先说评估。长运行、有状态的 agent 怎么测？Harbor + LangSmith 的组合给出一个 runner，能挂载 sandbox…

Langshaw: Declarative Interaction Protocols Based on Sayso and Conflict

Large and Deep Factor Models

Latent Actions from Factorized Transition Effects under Agent Ambiguity

Latent Bridges for Multi-Table Question Answering

Latent-CURE for Breast Cancer Diagnosis

Latent Noise Mask for Reducing Visual Redundancy in Multimodal Large Language Models

LatentRevise: Learning from Zero-Hit Reasoning

Lateral String Stability for Vehicle Platoons

LatSearch: Latent Reward-Guided Search for Faster Inference-Time Scaling in Video Diffusion

LaVPR: Benchmarking Language and Vision for Place Recognition

LC-ICL: Label-Guided Contrastive In-Context Learning for Robust Information Extraction

LeAD-M3D: Leveraging Asymmetric Distillation for Real-Time Monocular 3D Detection

Leader Reward for POMO-Based Neural Combinatorial Optimization

LeapXpert lands $180M to extract more intelligence from governed enterprise communications

Learned Coordination Conventions in Cooperative MARL: Measuring the Translation Gap Between Theory-Informed Roles and Learned Routing

Learned iterative networks: An operator learning perspective

Learning as Observable Matrix Dynamics: Diffusive Relaxations versus Phase Transitions

Learning Cross-view Correspondences for Geo-localization on Planetary Surfaces

Learning Deterministic Finite-State Machines from the Prefixes of a Single String is NP-Complete

Learning Efficient 4D Gaussian Representations from Monocular Videos with Flow Splatting

Learning from Acquisition: Metadata-driven Multimodal Pre-training for Cardiac MRI

Learning from Mistakes: Rollout-Retrieval Lifelong Policy Learning for Autonomous Driving

Learning from Reliable Latent Prompts for Visual Recognition with Missing Modalities

Learning from samples: inverse problems over measures

Learning How to Use Tools, Not Just When: Pattern-Aware Tool-Integrated Reasoning

Learning the structure of open quantum systems

Learning to Adaptively Allocate Gaussians for Arbitrary-Scale Image Super-Resolution

Learning to Bid in Discriminatory Auctions with Budget Constraints

Learning to Distributedly Estimate under Partially Known Dynamics: A Covariance-Agnostic Neural Kalman Consensus Filter

Learning to Route and Schedule LLMs from User Retrials via Contextual Queueing Bandits

Learning to Segment Liquids in Real-world Images

Learning Transferable Dynamics Priors from Action to World Modeling

LangChain 今天连发两记，直接把 agent reliability 往前推了一步：一套统一评估栈，一个不用完整沙箱的代码执行方案。先说评估。长运行、有状态的 agent 怎么测？Harbor + LangSmith 的组合给出一个 runner，能挂载 sandbox…