If you want to find the most pedantic corner of the internet, post a thought about AI on LinkedIn.
I recently found myself in a digital melee fight after using the blanket term ‘AI’ to describe LLMs. I suppose the responses were predictable:
“Usage of AI as a synonym for LLM tells me a lot about the AI knowledge of the author.”
Ouch, LOL.
So I shared a definition of Large Language Models (LLMs) from Microsoft’s own documentation:
“Large language models (LLMs) are advanced AI systems that understand and generate natural language, or human-like text, using the data they’ve been trained on through machine learning techniques.”
And the “well, actually” crowd came out in full force, arguing that calling an LLM “AI” is a logical fallacy, like saying a thumb is a hand. And while they aren’t technically wrong about the hierarchy, they’re missing the forest for the trees.
As someone who spent 20 years in the Canadian residential renovation business before pivoting into content creation and operations (where I work alongside LLMs and other GenAI tools), I don’t have much patience for “perfect systems” that don’t actually work. In construction, if the plan says “a beam goes there,” but the beam is actually six inches to the left, you don’t stand there arguing with the materials; you adjust.
Right now, the terminology around AI is the “misplaced beam” of the tech world. We’re arguing about whether a tool is an LLM, a Diffusion Model, or “True AI,” while the people we’re trying to help are just getting more overwhelmed.
We need a better mental model.
We need a taxonomy.
To understand why the LinkedIn experts are grumpy, we have to look at how these things are actually related. Think of it like biology. You wouldn’t call a house cat a “mammal” and expect people to know exactly what you’re talking about; it’s too broad. But a cat is a mammal.
Here is the “clean” taxonomy of the AI world that my wetware has conjured:
This is the big tent. It includes everything from the simple rule-based systems of the 1980s to the GPS on your phone. If a system performs a task we usually associate with “intelligence”—reasoning, perception, or decision-making—it lives in this kingdom.
This is a major subset of AI. Instead of being hand-coded with “If/Then” rules, these systems learn patterns from data.
This is where the current hype lives. While traditional ML might classify a photo of a cat, Generative AI creates a new photo of a cat that doesn’t exist. It generates novel content: text, images, audio, or video.
This is where we split the world by what the machine actually produces:
Text Generation, Image Generation, Video Generation, and Audio Generation
Now, this is the point of contention.
These are language-first models. They are trained to predict the next “token” (word or sub-word).
These are the engines behind tools like Midjourney or Sora. They work by “denoising”—taking a field of static and slowly refining it into a coherent image or video.
Finally, we get to the things you actually use, like ChatGPT, Claude, Gemini, Veo, and Runway.
The reason the LinkedIn experts get their high-visibility vests in a knot is that people use “AI” (the Kingdom) as a synonym for “LLM” (the Species).
Is it a “straightforward logical mistake,” as one of the LinkedIn commenters pointed out? Sure. But in common parlance, it’s just shorthand. When someone says, “I’m using AI to write this email,” they aren’t claiming they’ve achieved AGI (Artificial General Intelligence). They’re just skipping five rungs of the taxonomy for the sake of a sentence.
If we stopped there, the taxonomy would be perfect. But reality, especially in tech, is rarely that clean. The biggest source of confusion is that the “tools” we use aren’t single organisms. They’re more mini ecosystems; a blend of different features and models.
Or if you’re a bit quirky like me, you see them as Chimeras.
In mythology, a Chimera is a single creature made from the parts of different animals, think the Griffin, the Manticore or the Greek Sphinx.
Modern AI products are exactly like that. Take Google’s Veo 3 (a text-to-video generator) for example. If you ask, “Is Veo an LLM?” the answer is: “It’s complicated.”
As a core generator, Veo is a video foundation model (likely diffusion-based). It isn’t “predicting the next word.” However, the Veo system often includes an LLM “head.” The LLM is there to interpret your prompt, expand on it, and plan the “shots.”
So, it’s an LLM head on a Diffusion body.
This is why the “LLM isn’t AI” debate is so exhausting. Most tools you touch are actually symbiotic ecosystems. ChatGPT isn’t just an LLM anymore; it’s an LLM that can trigger a Python interpreter, a DALL-E image generator, and a web-search tool.
In my initial thoughts on this, I likened this to interspecies breeding, but quickly realized it’s not; it’s a group of different species working together to get a job done.
I’ve spent my life turning messy inputs into clean plans. Whether I was renovating a kitchen in or creating my SEO content workflow for a marketing agency client in LA, the goal is always the same: Reduce friction.
When you understand the taxonomy, the “AI overwhelm” starts to evaporate. You realize that you don’t need to learn “AI”; you need to understand which species of tool solves your specific friction point.
If you need to reason through a complex contract? Use an LLM species, like Claude.
If you need to visualize a new kitchen layout for a client? Use a Diffusion species, like Runway.
And, if you’re using a tool like Gemini or ChatGPT, understand that what you’re actually using is a Chimera, a Multimodal tool that does both.
If I were starting my AI adoption journey today, I wouldn’t spend much time worrying about the “Is it really AI?” debate; beyond reading this post, obvs. That is just a distraction for people who want to look smart in comment sections.
Instead, I’d focus on system-first operation.
Identify the Modality: What do you actually need to produce? Text? Video? Data analysis?
Develop prompt engineering fundamentals (for LLMs): how to structure instructions, provide context, and iterate effectively
Look for the Chimera: Don’t just look for a model; look for a product that integrates the right “parts.” A great prompt is useless if the model doesn’t have the “limbs” (like web search or code execution) to carry it out.
Understand model limitations (for all families): what each type of model is genuinely good at, and where it reliably fails
Think in Systems (for hybrid tools): understanding that most products are pipelines, not single models—and learning to troubleshoot accordingly
Build Repeatable Workflows: Stop treating AI like a magic wand and start treating it like a subcontractor. Give it clear instructions, understand its limitations (its “species traits”), and always perform QA.
Remember: The taxonomy helps you organize your knowledge. But the real skill is knowing how to work with the tools, regardless of what you call them.
The “LLM isn’t AI” debate persists because it offers both sides an emotional payoff. One side gets to feel precise and historically informed. The other side gets to feel vindicated that they weren’t “wrong,” they were “oversimplifying.” Then the debate loops, and nothing changes.
Meanwhile, the workplace reality is simple: people are using these systems—whatever you want to call them—to draft, summarise, plan, brainstorm, code, and compress time.
So the more useful question isn’t “Is an LLM AI?”
The useful question is: “Do you understand the ladder well enough to use the tool responsibly and evaluate it clearly?”
If you do, you’re ahead. If you don’t, you’ll keep losing time to semantics and tool-hopping.
So, the next time you see a digital bar fight on LinkedIn about whether an LLM is “true AI,” just remember: they’re arguing about the labels on the specimen jars while the rest of us are out here building with the zoo.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.