We are made of star-stuff is an inventory. We are a way for the cosmos to know itself is a different kind of claim. Separating the two.

If the universe were infinite and full of stars, the night sky would blaze. Olbers' paradox, its real history, and why dark means not yet.

Dreams, sleep paralysis, out-of-body experiences and the 21-gram story: what each one actually shows about the brain, and the one question that remains after all of them are explained.

From Earth as a pixel to a civilization as a few centimetres of rock: why smallness describes a scale, and why the same distance that shrinks everything also makes one place matter.

Light crosses the room instantly and the galaxy in 100,000 years. Why the speed of light is slow at cosmic scale, and what that says about the human ruler.

Game-playing AI tooling splits into environments, algorithms and agents. A map of the main projects in each layer, with Gymnasium, PettingZoo and CleanRL actually run as examples.

A hands-on walkthrough: train a PPO agent to play Atari Breakout from raw pixels in about 30 minutes without a GPU. Every command and number here comes from a real run.

Reinforcement learning from pixels, self-play at scale, imitation from video, and LLM agents that write code: four routes by which AI learns to play games, with a runnable example.

A large language model is a short list of mathematical operations composed many times: dot product, matrix projection, softmax, weighted averaging, and cross-entropy with gradient descent.

Choosing a model and judging your own application are two different evaluation problems. Benchmarks answer the first. Error analysis, graders and a small eval set answer the second.