Embedded Universal Predictive Intelligence: a coherent framework for multi-agent

NeurIPS 2025 Best Papers in Comics: From Artificial Hivemind to 1000-Layer RL

Visualizing Research: How I Use Gemini 3.0 to Turn Papers into Comics

Arc Is a Vision Problem

AlphaResearch: Accelerating New Algorithm Discovery with Language Models

LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics

Nested Learning: The Illusion of Deep Learning Architectures

Context Engineering 2.0: The Context of Context Engineering

A Practitioner's Guide to Kolmogorov-Arnold Networks

Kimi Linear: An Expressive, Efficient Attention Architecture

The Principles of Diffusion Models (470-pages)

CaT Replaces CoT-SC / Compute as Teacher: Turning Inference Compute Into

Tiny Recursive Model (TRM) vs. Hierarchical Reasoning Model (HRM)

Barbarians at the Gate: How AI Is Upending Systems Research

Evolution Strategies at Scale: LLM Fine-Tuning Beyond Reinforcement Learning

Autoreview: The Dragon Hatchling – The Missing Link Between the Transformer and

Stochastic Activations

LLM-JEPA: Large Language Models Meet Joint Embedding Predictive Architectures

Review: SpikingBrain Technical Spiking Brain-Inspired Large Models

K2-Think: A Parameter-Efficient Reasoning System

Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of AI

Fantastic Pretraining Optimizers and Where to Find Them

Solving the compute crisis with physics-based ASICs

Critiques of World Models

DeepConf: Scaling LLM reasoning with confidence, not just compute

V-JEPA 2: Scaling V-JEPA

Speed Always Wins: A Survey on Efficient Architectures for Large Language Models

Tversky Neural Networks

Einstein Fields: A Neural Perspective to Computational General Relativity

Tversky Neural Networks: Psychologically Plausible Deep Learning With

GEPA: Reflective prompt evolution can outperform reinforcement learning

Subliminal Learning: Language models transmit behavioral traits via hidden

AlphaGo Moment for Model Architecture Discovery

Paper FOMO and ICML 2025 Outstanding Papers

Early Signs of Steganographic Capabilities in Frontier LLMs

Musicality in Animals

Text-to-LoRA Enables On-the-Fly Model Adaptation

Quantum computing and artificial intelligence: status and perspectives

The Most Misunderstood Feature of the Sound

Darwin Gödel Machine

Are Deeper LLMs Smarter, or Just Longer?

Muon Optimizer Accelerates Grokking

ThoughtTerminator

Chain of Continuous Thought (Coconut)

Intuitive Physics Emergence in V-JEPA

Sound physics And basics of sound perception

BLT: Byte Latent Transformer

A Single 'Super Weight' Can Break Your Billion-Parameter Model

Jax Things to Watch for in 2025

Diffusion models are evolutionary algorithms

Make Softmax Great Again

Deep Learning Frameworks: The Fourth Pillar of Deep Learning Revolution

TextGrad: Automatic "Differentiation" via Text

Superconducting Supercomputers

Decoder-decoder architecture is coming

Chronos: Using Pretrained LLMs for Probabilistic Time Series Forecasting

Big Post About Big Context

Neural Network Diffusion

Thermodynamic AI is getting hotter

Training LLMs with AMD GPUs on Frontier Supercomputer

Beyond Chinchilla-Optimal Accounting for Inference in Language Model Scaling Law

Project CETI

GonzoML on Mamba and S6 (+previous post on S4)

Conway's Game of Life Is Omniperiodic

GonzoML on Gemini

Matryoshka Representation Learning

Mindstorms in Natural Language-Based Societies of Mind

The convolution empire strikes back

Sparse Universal Transformer

MemWalker: An alternative way for working with long documents using transformers

"Building Machines That Learn and Think Like People", 7 Years Later

Chain-of-Thought → Tree-of-Thought

Mortal Computers

Levanter – Legible, Scalable, Reproducible Foundation Models with Jax

LM-3 –- resurrecting the MIT CADR

The Annotated Diffusion Model

Road to text-guided image generation: DALL·E, CLIP, GLIDE, DALL·E 2 (unCLIP)

Self-replicating radiation-shield for deep-space exploration: Radiotrophic fungi

Revisiting ‘Powers of Ten’ – what we’ve learned about the Universe since 1977

The Future of Artificial Intelligence Is Self-Organizing and Self-Assembling

Foundation Models

Unsolved ML Safety Problems

Nx (Numerical Elixir) is now publicly available

Hardware for Deep Learning. Part 4: ASIC

JAX Ecosystem

Using JAX to accelerate our research

Game Plan: What AI Can Do for Football, and What Football Can Do for AI

NYPD deploys robot dog after woman shot during Brooklyn parking dispute

List of Animals by Number of Neurons

Natural Nuclear Fission Reactor

Brain2Word: Decoding Brain Activity for Language Generation