Welcome to Neuron by Neuron

Hello, and welcome to Neuron by Neuron. I’m Neill, and I’m excited and a bit trepidatious to be writing something and putting myself out into the world. I’ve been in the tech space professionally for - my goodness I feel old - about 20 years. Currently, I work as an AI Architect, and I’m writing this for a few reasons:

1. Everyone should have the opportunity to understand this transformative technology.
2. It helps me think through these concepts and solidify my understanding.
3. There seems to be a lot of hype and content around the “what” and less focus on the “how”. Maybe, in some small way, I can contribute to the pieces that I feel are important.

Before we delve (see what I did there AI detectors and internet sleuths? I just lost my big model smell) into the deeper mechanics, let’s start with a question that can be difficult to answer clearly: What is AI? It might seem obvious given the recent hype around Chatbots and LLMs, but defining AI can be more complicated than it might first seem - and getting the definition right is important for understanding everything else.

What is AI?

Everyone has heard the term AI - it’s in our phones, social media, and seemingly every other tech-related headline. But what exactly are we talking about? For our purposes, we’ll define it as creating a machine or program that can do things we normally ascribe to human intelligence. An example:

Imagine teaching a child how to identify dogs, you might start by pointing out different dogs: “See the golden retreiver? that’s a dog. The little pug over there? That’s a dog, too.” Over time, the child learns to generalize that knowledge and can recognize dogs of all shapes and sizes, even if they’ve never seen them before. They have learned a pattern.

AI, in its current state, works in a similar way. Instead of being programmed with rigid rules like “has four legs, hair, and barks, so it’s a dog,” modern AI systems learn from examples - a huge number of examples. For the dog identification task described, we would use the term supervised learning - we provide the correct answer for the AI and let the AI model adjust to align more closely with the correct answer the next time it sees something similar.

Artificial intelligence has made its way into everyday life, even for people who don’t think they are interacting with AI. Any time Netflix recommends a movie or Spotify recommends a song, there’s an AI recommender system. When you talk to Siri, AI is there translating your speech to text, understanding what you said, and producing an output. Your email spam filter? That’s some sort of AI or Machine Learning (ML) based system.

Deep learning is a specific type of AI, a subset of a broader area called Machine Learning. You can think about these concepts like a set of Russian nesting dolls (Matryoshka dolls). The largest doll would be AI, within it is a smaller doll representing Machine Learning and nested within Machine Learning is an even smaller doll, Deep Learning.

Why should you care about AI, ML, and Deep Learning?

AI has already changed the world and will continue changing the world for the foreseeable future, impacting every industry from health and finance to transportation, entertainment, and education. Understanding the basics is becoming increasingly important no matter what field you’re in. Plus, it’s super cool! In this newsletter, we’ll explore how machines can learn, make decisions, and even be creative. In the process, we might even learn a little about ourselves.

What's next

In the coming posts, we’ll explore the vast (and I do mean VAST) world of AI, including:

• Classical AI & Search: How computers solve puzzles, play games, and find optimal solutions - simple pathfinding to the algorithms that power GPS navigation
• Machine Learning Fundamentals: Understanding how machines learn from data, from basic statistical approaches to modern deep learning
• Neural Networks Deep Dive: We’ll start building at a single neuron and work our way to understanding complex architectures like CNN (for vision) and transformers (for modern LLMs)
• Reinforcement Learning: How an AI can learn through trial and error - powering AIs that win Chess and Go
• Natural Language Processing: How computers understand and generate human language
• Computer Vision: How machines learn to “see” and understand images and video
• Latest Breakthroughs: We’ll take some breaks to talk about the latest news when it’s relevant

I don't think this is going to be a linear or in any way organized series. It'll be me writing about whatever I'm thinking about at the moment with the topics above as a general outline.

A Personal Note

After two decades in tech, I’ve seen a number of technological shifts, but AI stands apart (it hits different, as the kids might say). It’s not just another tool or framework - it’s something closer to the internet. What excites me most is sharing this in a way that peels back some of the layers of the complexity but preserves all of the wonder I still feel every day.

In the next post, we’ll start with one of AI’s foundational challenges: teaching computers to solve puzzles and play games. No advanced math degree required. (though I might include some math for nerds) - just bring your curiosity and willingness to learn. We’ll focus on the core concepts that make AI possible and hopefully get at the what's really doing the work [insert picture of a building full of GPUs]. Maybe that inserted picture should be one of the earth turned into a giant GPU cluster. Or a Dyson sphere to power them all.