← Back to posts
Mar 20, 2026
Zero To Engineer: AI, ML, Deep Learning, LLMs
Stop Using These Words Interchangeably: Where every Friday we discuss something new to help you become a better developer
Happy Friday!
Every time someone says “we’re using AI” to describe their recommendation algorithm, I’m not sure if they actually know what they are talking about.
But, I get it. The terms are everywhere, used loosely by everyone from LinkedIn influencers to Fortune 500 press releases. But if you’re an engineer or CS student, getting these wrong in an interview or a technical conversation will cost you credibility fast.
So, let’s fix that!
--------------------------------------------
The misconception:
they’re all the same thing
--------------------------------------------
They’re not. They’re nested inside each other. Think Russian dolls. AI is the outermost layer. LLMs are the smallest doll, tucked inside everything else.
Here’s the actual hierarchy:
Artificial Intelligence is the broadest umbrella. Any system designed to simulate human-like intelligence falls here, including rule-based systems from the 1950s that had nothing to do with data or neural networks. Chess engines, recommendation filters, and spam detectors are all AI.
Machine Learning is a subset of AI. The key distinction is that ML systems learn patterns from data instead of following hand-coded rules. You stop writing the rules yourself. You feed the algorithm examples and let it find the patterns. Decision trees, linear regression, SVMs, all ML.
Deep Learning is a subset of ML. It specifically uses artificial neural networks with many layers (that’s where “deep” comes from). This is what unlocked image recognition, voice assistants, and translation at scale. The jump from traditional ML to deep learning was massive, but it’s still just a slice of the larger ML category.
Large Language Models are a specific application of deep learning. They are massive transformer-based neural networks trained on enormous amounts of text. GPT, Claude, Gemini, LLaMA. These are LLMs. They are not “AI” in some generic sense. They are a very specific, very recent technology sitting at the bottom of a long stack.
--------------------------------------------
The misconception:
LLMs are the pinnacle of AI
LLMs are impressive, but framing them as the destination misses the mark. They are one tool inside a vast field. Robotics, reinforcement learning, computer vision, and planning systems, none of those are LLMs. Knowing where LLMs actually sit in the hierarchy makes you a sharper thinker about what they can and cannot do.
--------------------------------------------
The misconception:
If it uses AI, it must use neural networks
A spam filter trained on keyword frequency is AI. It might not touch a neural network at all. When someone says “our product uses AI,” that tells you almost nothing about the underlying technology. Start asking what kind.
--------------------------------------------
Why this matters for you
You will be in rooms where these terms get thrown around carelessly. Knowing the actual distinctions lets you ask better questions and write cleaner technical documentation.
You got this!
Let’s Build It Beautifully,
Fab.
Every time someone says “we’re using AI” to describe their recommendation algorithm, I’m not sure if they actually know what they are talking about.
But, I get it. The terms are everywhere, used loosely by everyone from LinkedIn influencers to Fortune 500 press releases. But if you’re an engineer or CS student, getting these wrong in an interview or a technical conversation will cost you credibility fast.
So, let’s fix that!
--------------------------------------------
The misconception:
they’re all the same thing
--------------------------------------------
They’re not. They’re nested inside each other. Think Russian dolls. AI is the outermost layer. LLMs are the smallest doll, tucked inside everything else.
Here’s the actual hierarchy:
Artificial Intelligence is the broadest umbrella. Any system designed to simulate human-like intelligence falls here, including rule-based systems from the 1950s that had nothing to do with data or neural networks. Chess engines, recommendation filters, and spam detectors are all AI.
Machine Learning is a subset of AI. The key distinction is that ML systems learn patterns from data instead of following hand-coded rules. You stop writing the rules yourself. You feed the algorithm examples and let it find the patterns. Decision trees, linear regression, SVMs, all ML.
Deep Learning is a subset of ML. It specifically uses artificial neural networks with many layers (that’s where “deep” comes from). This is what unlocked image recognition, voice assistants, and translation at scale. The jump from traditional ML to deep learning was massive, but it’s still just a slice of the larger ML category.
Large Language Models are a specific application of deep learning. They are massive transformer-based neural networks trained on enormous amounts of text. GPT, Claude, Gemini, LLaMA. These are LLMs. They are not “AI” in some generic sense. They are a very specific, very recent technology sitting at the bottom of a long stack.
--------------------------------------------
The misconception:
LLMs are the pinnacle of AI
LLMs are impressive, but framing them as the destination misses the mark. They are one tool inside a vast field. Robotics, reinforcement learning, computer vision, and planning systems, none of those are LLMs. Knowing where LLMs actually sit in the hierarchy makes you a sharper thinker about what they can and cannot do.
--------------------------------------------
The misconception:
If it uses AI, it must use neural networks
A spam filter trained on keyword frequency is AI. It might not touch a neural network at all. When someone says “our product uses AI,” that tells you almost nothing about the underlying technology. Start asking what kind.
--------------------------------------------
Why this matters for you
You will be in rooms where these terms get thrown around carelessly. Knowing the actual distinctions lets you ask better questions and write cleaner technical documentation.
You got this!
Let’s Build It Beautifully,
Fab.