AI Wiki — Knowledge Base

LangChain 在 Interrupt 大会上发布了底层数据库 SmithDB 和自动化排障引擎 LangSmith Engine。 Agent 运行会产生海量 trace（执行轨迹），把旧数据库撑到了瓶颈。新底座 SmithDB 放弃了本地磁盘，全面转向对象存储，将核心查询…

LangChain 在 Interrupt 大会上发布了底层数据库 SmithDB 和自动化排障引擎 LangSmith Engine。 Agent 运行会产生海量 trace（执行轨迹），把旧数据库撑到了瓶颈。新底座 SmithDB 放弃了本地磁盘，全面转向对象存储，将核心查询速度拉高了 15 倍。底座换新后，LangSmith Engine 顺势接管了查 Bug 的体力活。它在后台持续监控生

model-releasesharrison-chase--x

14 May 2026

All entries

Judge probes whether Musk settlement with Trump admin is tainted by corruption

Just Ask for a Table: A Thirty-Token User Prompt Defeats Sponsored Recommendations in Twelve LLMs

JUST IN: We're launching LangChain Labs. A new applied research effort focused on Continual Learning.

KamonBench: A Grammar-Based Dataset for Evaluating Compositional Factor Recovery in Vision-Language Models

KAST-BAR: Knowledge-Anchored Semantically-Dynamic Topology Brain Autoregressive Modeling for Universal Neural Interpretation

Kernel-based guarantees for nonlinear parametric models in Bayesian optimization

Kernel Embeddings and the Separation of Measure Phenomenon

Kimi K2.6 and DeepSeek V4 Pro are now GA on @FireworksAI_HQ on Foundry + PTU support in the US Data Zone—predictable performance, data resid…

Kimi K2.6 is now open-weight #1 on Finance Agent Benchmark V2.

Kiwi-Edit: Versatile Video Editing via Instruction and Reference Guidance

KVServe: Service-Aware KV Cache Compression for Communication-Efficient Disaggregated LLM Serving

LangChain 在 Interrupt 大会上发布了底层数据库 SmithDB 和自动化排障引擎 LangSmith Engine。 Agent 运行会产生海量 trace（执行轨迹），把旧数据库撑到了瓶颈。新底座 SmithDB 放弃了本地磁盘，全面转向对象存储，将核心查询…

LangChain Labs is live

LangSmith Engine, 6 months later

LangSmith Engine feels like the missing piece for reliable agents. Clustering failures and suggesting targeted fixes turns raw traces into a…

LangSmith Engine is a phase shift because traces are no longer just records to be manually inspected, they’re now the catalyst for recursive…

LangSmith Fleet now has a free model powered by @FireworksAI_HQ for Developer and Plus plans. It’s now easier than ever to get started. Try …

Language-Based Agent Control

Language Model Goal Selection Differs from Humans' in a Self-Directed Learning Task

Language Model Networks: Supervision-Efficient Learning through Dense Communication

Large Language Models for Agentic NetOps and AIOps: Architectures, Evaluation, and Safety

Large Language Models Lack Temporal Awareness of Medical Knowledge

Latent-Augmented Discrete Diffusion Models

Layer-wise Representation Dynamics: An Empirical Investigation Across Embedders and Base LLMs

LeanSearch v2: Global Premise Retrieval for Lean 4 Theorem Proving

Learning a Continue-Thinking Token for Enhanced Test-Time Scaling

Learning Local Constraints for Reinforcement-Learned Content Generators

Learning Perturbations to Extrapolate Your LLM

Learning POMDP World Models from Observations with Language-Model Priors

Learning Responsibility-Attributed Adversarial Scenarios for Testing Autonomous Vehicles

Learning to Approximate Uniform Facility Location via Graph Neural Networks

Learning to Decide with AI Assistance under Human-Alignment

Learning to Optimize Radiotherapy Plans via Fluence Maps Diffusion Model Generation and LSTM-based Optimization

Learning to See What You Need: Gaze Attention for Multimodal Large Language Models

Learning Transferable Latent User Preferences for Human-Aligned Decision Making

Learning When to Act: Communication-Efficient Reinforcement Learning via Run-Time Assurance

Learning with Rare Success but Rich Feedback via Reflection-Enhanced Self-Distillation

LENS: Multi-level Evaluation of Multimodal Reasoning with Large Language Models

LEXI-SG: Monocular 3D Scene Graph Mapping with Room-Guided Feed-Forward Reconstruction

Lifelong Learning in Vision-Language Models: Enhanced EWC with Cross-Modal Knowledge Retention

LIFT: Last-Mile Fine-Tuning for Table Explicitation

LightSplit: Practical Privacy-Preserving Split Learning via Orthogonal Projections

Like any AI dev not employed by a closed lab, we share the ambition that at least 10x more should be capable of training frontier models in …

Limitations of Quantum Advantage in Unsupervised Machine Learning

Limits of Personalizing Differential Privacy Budgets

LLM Flow Processes for Text-Conditioned Regression

LLMs as annotators of credibility assessment in Danish asylum decisions: evaluating classification performance and errors beyond aggregated metrics

LLMs as Implicit Imputers: Uncertainty Should Scale with Missing Information