At a recent conference, I found myself in the middle of a familiar conversation. A panelist was describing an assignment in which students used “ChatGPT.” Another speaker referenced a colleague who had banned “ChatGPT” from the classroom. Someone else asked whether “ChatGPT” could grade essays or generate lesson plans.
I counted the word six times in five minutes.
And not once did anyone refer to a specific model. In fact, the assignment being discussed didn’t even use ChatGPT. No one was talking about the same thing. No one meant the same capabilities. No one realized we weren’t even dealing with the same category of system.
As a researcher who studies digital spaces, tools, and literacies, this is the part that alarms me the most. We are building pedagogies, policies, and moral panics on terminology that has blurred into meaninglessness.
In public discourse, “ChatGPT” has become a shorthand term for anything that feels algorithmic, uncanny, automated, or AI-driven. However, if we want to understand how these systems shape reading, writing, learning, and human judgment, we need something stronger than vague notions and umbrella terms.
We need vocabulary. Even when the models themselves refuse to give it to us.
The Problem: We Use One Word for a Whole Ecosystem
Imagine if every computer were “Microsoft,” and every phone were an “iPhone.” Imagine referring to every search engine simply as “Google.” It would be like calling every photocopy a “Xerox.”
Actually… wait a minute. 🙂
That’s where we are with AI. In everyday discourse:
- ChatGPT refers to any generative text tool.
- AI sometimes means “machine learning,” sometimes “automation,” sometimes “autocomplete,” and sometimes “a spooky consciousness emerging inside my laptop.”
- ” The AI” is treated as a single entity rather than a fragmented, rapidly evolving set of architectures, models, reasoning engines, APIs, guardrails, and data pipelines.
This ambiguity isn’t only sloppy. It actively blocks understanding. It flattens the differences that matter for equity, access, pedagogy, ethics, and research methods.
The Hard Part: Even Experts Can’t Name What They’re Using
Models don’t know what they are. Interfaces don’t always tell you. Companies swap models in the background to reduce cost or meet demand. Training data cutoffs mean a model literally cannot know the existence of its own newer sibling.
So we end up here:
- We should distinguish between GPT-4 and GPT-4o.
- We should be able to identify when someone is using Claude 3.5 Sonnet instead of Gemini Flash.
- We should be able to ask “what model is this?” and get an accurate answer.
But we often can’t.
And yet, as educators and researchers, we cannot throw up our hands and say, “Well, it’s all just AI anyway.” The differences remain real. Even when the category labels are unstable.
Why It Matters: Capabilities, Limits, and Misconceptions
If we collapse everything into “AI,” we lose the ability to ask the right questions.
- Recognizing different strengths. Some models excel at coding, while others excel at reasoning, summarizing, and multimodal analysis.
- Anticipating different weaknesses. Some hallucinate worse than others. Some mis-handle logic puzzles. Some fail at counting letters. Some individuals are unable to evaluate citations or code.
- Designing appropriate learning tasks. A prompt that works beautifully on a large reasoning model might fail completely on a small, fast model.
- Understanding risk. A local LLM running on a homelab server has vastly different privacy implications than a cloud model collecting telemetry at scale.
- Engaging ethically. Research norms, informed consent, data privacy, and accessibility all hinge on knowing what system is doing what work.
If we treat these systems as if they are interchangeable, then we stop asking the questions that help us make informed decisions.
The Digital Literacy Angle: Naming Is Power
In digital literacy research, we teach students to identify sources, examine provenance, understand the architecture of tools, and interrogate the terms we casually use. This moment in AI is no different.
If we want students, teachers, and policymakers to interact responsibly with AI systems, they need enough technical literacy to distinguish:
- Platform vs. Model
- Model vs. Product
- AI vs. Automation
- Generative system vs. Retrieval system
- Reasoning engine vs. Pattern generator
- Hallucination vs. Uncertainty vs. Fabrication
The words we choose shape the questions we ask. The questions we ask shape the policies we create. And the policies we create shape the futures we walk into.
So What Do We Do?
We probably can’t fix the fact that models won’t reliably tell you who they are. We probably can’t stop companies from swapping models under the hood. We probably can’t expect the general public to differentiate transformer architectures from fine-tuning runs.
But we can do something.
- Teach the difference between tools rather than flattening them into one word.
- Build assignments that require students to identify the model, platform, and (if applicable) version.
- Encourage educators to stop saying “ChatGPT” when they mean “any generative AI tool.” If you start referring to everything as “Generative AI,” you’re better off.
- Push for transparency in platforms that hide model identities.
- Help our research communities articulate clearer language…not perfect, but better.
Even if it’s imperfect, a shared vocabulary is the beginning of shared understanding.
Closing Thoughts
We live in a moment when we have both too many names for AI and not nearly enough. The public often imagines a single, monolithic “AI,” while researchers encounter hundreds of different models, each with its own capabilities, failures, quirks, and harm profiles.
If we want to build literate, critical, humane, and ethical engagements with these systems, we must start by teaching ourselves and our communities how to name what we are dealing with.
Even when the models themselves can’t.