Lattner: Compilers, Swift, and Democratizing AI

Resources

Resources & Recommendations

Books

• "Beyond Vibe Coding" by Addy Osmani - This book discusses how to enable a verbose output for AI agents and understanding the generated code to maintain human oversight.

Research & Studies

• "Attention Is All You Need" - This paper, developed on TPUs, is a foundational work in AI, particularly for transformer models.

Tools & Software

• LLVM - A collection of modular and reusable compiler and toolchain technologies, used by languages like Swift, Rust, and C. It started as a code generation system and is now broadly adopted.
• MLIR - A compiler infrastructure for machine learning that acts as a "2.0" version of LLVM for domain-specific chips in AI.
• Clang - A C language family frontend to LLVM, which was a significant part of replacing Apple's developer tools.
• XLA - A domain-specific compiler for linear algebra that optimizes TensorFlow computations, developed at Google for TPUs.
• Metal - Apple's low-level, low-overhead 3D graphics and compute API, part of their ML stack.
• MLX - Apple's machine learning framework, part of their ML stack.
• ROCm - AMD's open software platform for GPU computing, similar to CUDA.
• CUDA - NVIDIA's parallel computing platform and programming model for GPUs, widely used in AI.
• TensorFlow - An open-source machine learning framework.
• PyTorch - An open-source machine learning framework.
• ONNX Runtime - A cross-platform inference engine for machine learning models.
• Cafe - An early deep learning framework, mentioned in the context of Chris Lattner's work at Tesla.
• Autoconf - A macro processing tool used in the past to work around compiler limitations.
• Cursor - An AI coding tool used by Chris Lattner for daily coding, providing productivity benefits for mechanical rewrites.
• Claude Code - An AI coding tool.
• Xcode - Apple's integrated development environment, used for Swift development.
• Xcode Playgrounds - An interactive environment within Xcode for experimenting with Swift code.
• REPL (Read-Eval-Print Loop) - An interactive programming environment available for Swift.
• SwiftUI - Apple's declarative UI framework, which helped drive further adoption of Swift.
• Zig - A programming language known for its compile-time metaprogramming capabilities, which influenced Mojo.

People Mentioned

• Richard Stallman - Founder of the GNU Project, mentioned as being opposed to C.
• Vikram Adve - Chris Lattner's university advisor who encouraged the development of LLVM.
• Steve McConnell - Author, mentioned for his insights on good software development practices, particularly writing code multiple times.
• Addy Osmani (Chrome DevTools team) - Mentioned for his book "Beyond Vibe Coding" and approach to using AI tools.

Organizations & Institutions

• Apple - The company where Chris Lattner developed LLVM, Clang, and Swift.
• Uber - Mobile-first company that adopted Swift for its app rewrite.
• Cray - Built a supercomputer using LLVM.
• Google - Adopted LLVM and where Chris Lattner worked on TensorFlow and TPUs.
• Intel - Company that eventually canceled internal compilers and switched to LLVM.
• ARM - Company that eventually canceled internal compilers and switched to LLVM.
• Anthropic - AI company mentioned for their engineering blog post about reimplementing models for different hardware.
• SciFive - Builds RISC-V hardware, where Chris Lattner worked on chip design and AI IP.
• Modular - Chris Lattner's current company, focused on AI software and the Mojo programming language.

Courses & Educational Resources

• Kaleidoscope Tutorial (LLVM) - A tutorial written by Chris Lattner to teach about compilers using LLVM.

Websites & Online Resources

• Mojo Website - Contains tutorials, including "GPU puzzles," for learning the Mojo programming language.
• Anthropic Engineering Blog Post - Discusses the challenges of reimplementing AI models for different hardware (link to be provided in show notes).
• GitHub - Mentioned as a place to view Chris Lattner's historical code contributions, including to Swift.