The AI Timeline

How A Vision Transformer Just Broke The ARC-AGI Benchmark

Plus more on Seer, Virtual Width Networks, SAM 3, and Evolution Strategies at the Hyperscale

Depth Anything 3: Recovering the Visual Space from Any Views

LeJEPA, The Path Not Taken, and more

From Memorization to Reasoning in the Spectrum of Loss Curvature

Continuous Autoregressive Language Models and Introducing Nested Learning: A new ML paradigm for continual learning

Language Models are Injective and Hence Invertible

and more on Kimi Linear, Looped Transformer, How FP16 fixes RL...

How to Compress Long Text into Images To Reduce LLM Tokens

Last Week's Trending Papers 📈

RAE, Reasoning with Sampling, and more

Less is More: Recursive Reasoning with Tiny Networks

Plus more about Moloch's Bargain: Emergent Misalignment When LLMs Compete for Audiences and LLM Fine-Tuning Beyond Reinforcement Learning

Training Agents Inside of Scalable World Models

Plus more about Polychromic Objectives for Reinforcement Learning and Stochastic activations

Video models are zero-shot learners and reasoners

Plus more about Thinking Augmented Pre-training and Reinforcement Learning on Pre-Training Data

Pre-training under infinite compute

Plus more about Discovery of Unstable Singularities and AToken: A Unified Tokenizer for Vision

Defeating Nondeterminism in LLM Inference

Plus more about Analog in-memory computing attention mechanism for fast and energy-efficient large language models, and the Majority is not always right: RL training for solution aggregation

Why Do MLLMs Struggle with Spatial Understanding?

Plus more about Small Language Models are the Future of Agentic AI and Why Online Reinforcement Learning Forgets Less