TheSequence

Your agent got better this quarter and nobody trained anything. Where did the improvement go? Into the notes. This is the loop that runs between model releases, and it might be where the durable value

Smarter models, wearable agents, scientific discoveries, and superpower diplomacy reveal how quickly AI is moving into the world.

How text diffusion models rethink language generation—and the tradeoffs that will determine their future.

Three ways to turn model intelligence into working software

Inside the RSI post-training pipeline

Google tackles real-time interaction, OpenAI targets legal workflows, Figure tests unfamiliar homes, and investors double down on compute.

How Chinese and American labs pursue the next generation of intelligence

How architecture and system design turn model capability into useful work

Three frontier labs have now said out loud that their models help build their models. The interesting part is not the percentage. It is which half of the job got automated, and why.

Faster models, nine billion genetic predictions, personal agents, and a proof that could make mathematical history.

Generalist models are expanding what robots can do. The real breakthrough will be how easily we can teach them something new.

From Berkeley’s Chatbot Arena to Agent Arena: preference rankings, cost-per-task frontiers, and the hard problems in measuring real-world AI utility.

Three releases that deserve your attention.

From DistilBERT and Gemini Flash to Gemma, Llama, Qwen, DeepSeek, Phi, Ministral, and PrismML’s Bonsai 27B.

NVIDIA, Anthropic, NScale, and a16z show how the AI race is moving from models to infrastructure.

Robotics Gets Its PyTorch Stack.

Jensen Huang’s five-layer cake explains how intelligence is manufactured. The missing layer explains how quickly - and by whom - it can scale.

Distilling three major AI releases to keep you current.

The Sequence — Distillation Series

OpenRouter, Ramp, Etched, and DeepSeek reveal the emerging economic stack beneath modern intelligence.

Why the next frontier in intelligence will be constrained not only by algorithms and chips, but by megawatts, transmission lines, and the physics of heat.

A mini deep dive into some of the most important AI releases of last week.

Why test-time compute distillation could turn inference-time reasoning into permanent model capability.

Deep diving into three major AI releases.

NVIDIA’s new Nemotron 3.5 Lightning, other Nemotron models, architectures and more.

Compressing Time, Space, and Alignment

Jeff Dean leaves, Demis Hassabis moves upstream, and Muse Code turns software development into an orchestration problem.

Token maxing was the adoption phase. Intelligence resource planning is what comes next.

Google put a walking humanoid inside a single policy, published the reasoning half as an API, and kept the motor half behind a partner gate. The numbers explain why.

Weird but more common than you think. The type of distillation you were not thinking about.