optimization

Casey Muratori's first refterm lecture lays out philosophies of optimization, distinguishing optimization from non-pessimization, which means avoiding needless work. It uses his fast terminal renderer project to show how fast computers really are.

The introductory lecture of Stephen Boyd's Stanford Convex Optimization I course (EE 364A). It outlines convex sets, functions and problems, and the course's coverage of least-squares, linear and semidefinite programming, duality and interior-point methods.

Casey Muratori's University of Twente guest lecture is a case study in how straightforward code can cut thousands of lines and tens of seconds of runtime down to a few dozen lines running in under a second. It uses concrete examples to show why software is often an order of magnitude slower than it should be.

A guest lecture Jon Gjengset gave at Two Sigma on using Rust for large, high-performance systems. It covers why Rust suits such projects, the Noria research database, a concurrent data structure used in it, and practical experience.

Casey Muratori's invited lecture on where bad code comes from, rebroadcast from a Twitch stream. He critiques conventional wisdom such as SOLID and Clean Code and argues about how those habits lead to slow, poorly structured software.

Andrej Karpathy's first Neural Networks: Zero to Hero lecture, which builds micrograd, a tiny autograd engine, from scratch in Python. It explains backpropagation and neural network training step by step, assuming only basic Python and some high-school calculus.

Andrej Karpathy's three-hour-plus general-audience deep dive into how ChatGPT-style LLMs are built, covering the full training stack from pretraining to post-training. It also gives mental models for their quirks, like hallucination, and how to use them well.

Andrej Karpathy builds the GPT tokenizer from scratch, implementing byte pair encoding along with encode and decode. He shows how many odd LLM behaviors trace back to tokenization, and covers Unicode byte encodings along the way.

Leo Dirac explains how transformer models have replaced LSTMs in NLP. He gives background from bag of words through RNNs and LSTMs, then a technical deep dive into multi-headed self-attention and positional encoding, with sample code.

Ian Piumarta's 2007 Stanford EE380 lecture on building a dynamic programming environment from the ground up, from the Viewpoints Research Institute's work. It covers the design of a language and system with a recursive design, parsing through code generation, and the COLA architecture.

Anders Hejlsberg whiteboards how compilers are taught versus how modern ones are actually built. He explains the shift to compilers as services with incremental, error-tolerant, IDE-oriented design, as in Roslyn and TypeScript.

Geoffrey Hinton's 2025 Royal Institution Discourse on what AI has learned from biological intelligence and how it might surpass it. He explains neural networks and argues that human language use resembles what backpropagation-trained networks do.

The first lecture of Stanford's Natural Language Processing with Deep Learning course, taught by Chris Manning and Richard Socher. It introduces NLP and word vectors, including SVD, skip-gram, CBOW, negative sampling and word2vec.

Sasha Rush surveys linear RNNs and state space models as replacements for attention in transformer-style models, in a talk for MLSys. It covers the foundational preliminaries and predates Mamba.

Peter Norvig's UBC distinguished lecture on how massive amounts of simple data, such as for language modeling, can yield understanding that hand-built models could not.

The first lecture of David MacKay's Cambridge course on information theory, pattern recognition and neural networks. It introduces the core ideas of his book Information Theory, Inference, and Learning Algorithms.

A university lecture from CS480/680 by Pascal Poupart on attention mechanisms and transformer networks, worked through on a chalkboard.

John Gustafson presents posits, a proposed drop-in replacement for IEEE 754 floats that is simpler in hardware and more flexible in precision. The seminar includes a live demo and machine learning results.

Sacha Arnoud, Waymo's Director of Engineering, gives a lecture in MIT's 2018 deep learning for self-driving cars course on the rise of machine learning in autonomous vehicles. It covers how Waymo uses deep learning and how it handles tricky driving situations.

Bill Dally of NVIDIA gives a University of Washington lecture on trends in deep learning hardware. It covers numeric formats, specialized instructions, energy per operation and co-design of hardware and software.

Geoffrey Hinton gives an MIT talk on what is wrong with convolutional neural networks and introduces his capsules project. The description points to the capsule network papers.

The first lecture of Stanford's CS231n course on convolutional neural networks for visual recognition. It surveys the history and key challenges of computer vision, from the Cambrian explosion and Hubel and Wiesel to ImageNet.

Lecture 7A from MIT 6.001 (SICP) by Abelson and Sussman introduces the metacircular evaluator, a Scheme interpreter written in Scheme. It shows how evaluating a program is itself a program.

Lecture 1B of MIT's 6.001 Structure and Interpretation of Computer Programs (1986) with Gerald Sussman, covering procedures, processes and the substitution model. It includes the line about the power of naming things.

A long Mill Computing lecture on the Mill CPU's compiler toolchain, covering the specializer that targets family members, scheduling and latency passes, and the belt. It explains how the compiler handles a wide-issue, exposed-pipeline design.

Lecture 7A of the 1986 MIT 6.001 SICP course, with Gerald Sussman presenting the metacircular evaluator, part 1. It shows how a Lisp evaluator can be written in Lisp itself.