AI Wiki — Knowledge Base

全网都在吹的LeCun新论文，90%的解读都是错的。他们说生成式AI是死路，说过去三年花的几百亿全白费了，说15M参数的小模型就能吊打万亿大模型。这些全是营销号的夸张，我觉得这篇论文的真正分量比他们吹的还要重。 Yann LeCun团队这次解决了JEPA困扰了好几年的表征坍…

全网都在吹的LeCun新论文，90%的解读都是错的。他们说生成式AI是死路，说过去三年花的几百亿全白费了，说15M参数的小模型就能吊打万亿大模型。这些全是营销号的夸张，我觉得这篇论文的真正分量比他们吹的还要重。 Yann LeCun团队这次解决了JEPA困扰了好几年的表征坍缩问题。以前的世界模型，学着学着就会把狗车人都压成一模一样的向量，什么都学不到。这次他们只加了一个极其优雅的数学正则

researchyann-lecun--x

21 Apr 2026

All entries

Language Models Don't Know What You Want: Evaluating Personalization in Deep Research Needs Real Users

Large Language Models Are Bad Dice Players: LLMs Struggle to Generate Random Numbers from Statistical Distributions

Large Language Models Are Still Misled by Simple Bias Ensembles

LASER: Low-Rank Activation SVD for Efficient Recursion

Last Wednesday, our founder and scientific advisor, @Yoshua_Bengio, was officially appointed an Officer of the Order of the British Empire (…

Late Fusion Neural Operators for Extrapolation Across Parameter Space in Partial Differential Equations

Latent Abstraction for Retrieval-Augmented Generation

Latent-Compressed Variational Autoencoder for Video Diffusion Models

Latent Phase-Shift Rollback: Inference-Time Error Correction via Residual Stream Monitoring and KV-Cache Steering

Latent Preference Modeling for Cross-Session Personalized Tool Calling

LatentMimic: Terrain-Adaptive Locomotion via Latent Space Imitation

LayerCache: Exploiting Layer-wise Velocity Heterogeneity for Efficient Flow Matching Inference

LBFTI: Layer-Based Facial Template Inversion for Identity-Preserving Fine-Grained Face Reconstruction

LEAF: Knowledge Distillation of Text Embedding Models with Teacher-Aligned Representations

Learned Nonlocal Feature Matching and Filtering for RAW Image Denoising

Learning-Based Sparsification of Dynamic Graphs in Robotic Exploration Algorithms

Learning from Less: Measuring the Effectiveness of RLVR in Low Data and Compute Regimes

Learning Invariant Modality Representation for Robust Multimodal Learning from a Causal Inference Perspective

Learning residue level protein dynamics with multiscale Gaussians

Learning Stable Predictors from Weak Supervision under Distribution Shift

Learning the Riccati solution operator for time-varying LQR via Deep Operator Networks

Learning to Control Summaries with Score Ranking

Learning to Correct: Calibrated Reinforcement Learning for Multi-Attempt Chain-of-Thought

Learning to Retrieve User History and Generate User Profiles for Personalized Persuasiveness Prediction

Learning to Seek Help: Dynamic Collaboration Between Small and Large Language Models

Learning to Trade Like an Expert: Cognitive Fine-Tuning for Stable Financial Reasoning in Language Models

Learning Unanimously Acceptable Lotteries via Queries

Learning Whole-Body Humanoid Locomotion via Motion Generation and Motion Tracking

LEPO: nderline{L}atent Rnderline{e}asoning nderline{P}olicy nderline{O}ptimization for Large Language~Models

🤩 LES LENSOIS SONT COMME DANS UN RÊVE !!!!! 𝗘𝗡𝗩𝗔𝗛𝗜𝗦𝗦𝗘𝗠𝗘𝗡𝗧 𝗗𝗘 𝗧𝗘𝗥𝗥𝗔𝗜𝗡 𝗔̀ 𝗕𝗢𝗟𝗟𝗔𝗘𝗥𝗧 !!! ❤️💛 🎥 @francetvsport

Less Noise, More Voice: Reinforcement Learning for Reasoning via Instruction Purification

Let's talk parsing charts 📊📈. Last week we released ParseBench, the first document OCR benchmark for AI agents. New in ParseBench: ChartDa…

Leveraging Kernel Symmetry for Joint Compression and Error Mitigation in Edge Model Transfer

Leveraging Large Language Models for Sarcastic Speech Annotation in Sarcasm Detection

Leveraging VR Robot Games to Facilitate Data Collection for Embodied Intelligence Tasks

LexRel: Benchmarking Legal Relation Extraction for Chinese Civil Cases

LiDAR-based Crowd Navigation with Visible Edge Group Representation

LiFT: Does Instruction Fine-Tuning Improve In-Context Learning for Longitudinal Modelling by Large Language Models?

LIFT the Veil for the Truth: Principal Weights Emerge after Rank Reduction for Reasoning-Focused Supervised Fine-Tuning

Light-Adapted Electroretinogram and Oscillatory Potentials (LEOPs) Dataset for Autism Spectrum Disorder and Typically Developing Individuals

Lightweight Cybersickness Detection based on User-Specific Eye and Head Tracking Data in Virtual Reality

Lil: Less is Less When Applying Post-Training Sparse-Attention Algorithms in Long-Decode Stage

Linear-Time and Constant-Memory Text Embeddings Based on Recurrent Language Models

Linking Exteroception and Proprioception through Improved Contact Modeling for Soft Growing Robots

LiquidTAD: An Efficient Method for Temporal Action Detection via Liquid Neural Dynamics

Live Avatar: Streaming Real-time Audio-Driven Avatar Generation with Infinite Length

LIVE: Leveraging Image Manipulation Priors for Instruction-based Video Editing