This is not your typical AI article.

I'm not going to teach you model-specific tips that will last until the milk in your fridge expires. I'm interested in highlighting the four most common mistakes I see college freshmen make using AI. Whatever changes come in the space, whatever model you use, whatever task you are doing, fixing these mistakes will always make your output better.

These mistakes are not the obvious "prompt it better." You know that. I'm going to highlight the mistakes you don't even realize are mistakes, because they operate beneath the shadows of awareness.

Mistake 1: You Don't Know Yourself Well Enough

I still remember when ChatGPT first came out in November 2022. I was sitting in Olin Library, drinking coffee (black like my soul), when I jokingly asked it to write me an article intro about the power of meta-learning. I'd have been impressed if it could write correctly in english and not Hindu. I started to get up for a walk, thought it would take a few minutes. But holy shit. There was the intro, its white text arrayed in neat paragraphs like it was mocking me. It wasn't half bad either.

Over the next few months I began leaning on AI more and more, for both classes and blog posts. It started as a nudge here and there: rewriting a sentence, generating a hook. Then it became the first thing I opened every time I sat down to write. Then one afternoon at Olin, I opened ChatGPT before I even had an outline. I was going to let it write the whole thing.

I gazed out the window, stunned. What the hell was I doing? Didn't I want to improve my writing skills? Yet somehow I had unconsciously hired AI as my personal ghostwriter, and I hadn't even posted the job listing.

The problem wasn't that I was using AI. The problem was that I had no clarity on why I was using it, where it should help, and where it should stay out of my way. I didn't know myself well enough to draw those lines.

Without self-understanding, AI becomes a seed with no sunlight. It has no idea where to grow.

What overarching purpose are we using AI for? What values should drive that use? How should we use AI differently from every other student copying and pasting the same prompt into the same chatbot at 2am? This lack of self-knowledge creates two problems that most students never see coming.

The first problem is output monoculture.

When every student uses AI the same way, we get the same sort of output. AI is trained on the sum of human knowledge, and most of that sum is not that developed. It flattens developmental depth down to whatever the training data averaged out to: Western, liberal, educated, and Steve Jobs.

In other words, if AI was a human, it would be a white guy who peaked in a consulting internship. If you want outputs that carry the depth of higher stages of ego development (check out my article on mapping ego development to learn more), you yourself must grow into those stages. The AI can't take you somewhere you haven't been.

The second problem runs deeper: Without self-understanding, we use AI as an emotional pacifier.

This is why students replace their learning with AI. It's not just because it's useful. It's because we don't have the vision for our life that helps us push through discomfort. That blank page, that problem set that makes your brain itch, that essay question you genuinely don't know how to answer. AI removes the discomfort.

And if feels sooooo good.

This genuinely terrifies me. Imagine a cohort of students graduating college who have grades which indicate learning. But it's not them who received the diploma, our God, ChatGPT did.

When I get stuck in my writing now, what stops me from handing the whole thing to Claude isn't discipline. It's reminding myself of the sorry freshmen who will be helped by my articles, the ones reading this right now who need to hear something real, not something generated. Who need someone that continues to grow their consciousness as they help freshmen grow theirs.

Vision is the antidote to avoidance.

So what do you actually do about this?

First, journal.

Not in the "dear diary" sense. Define your values. Clarify your purpose with something like my Cosmic Journaling Kit. Then ask yourself a deep question where do I want to be a cyborg, and where do I want to be a cyclops?

Being a cyborg means using AI as a colleague in the places where you decide it's worth it. Being a cyclops means keeping your one good eye sharp in the places you value the human touch. Maybe you let AI help you brainstorm but you write the first draft alone. Maybe you let it organize research but you form your own arguments. The line is yours to draw, but you have to draw it consciously, or it gets drawn for you by convenience.

Second, train AI on yourself.

This is what breaks the monoculture. When AI knows your voice, your values, your preferences, and your blind spots, the output stops sounding like everyone else's and starts sounding like you. I've trained mine on my Obsidian notes, my preferred writing voice (including a whole document on what not to do), my book list, my major articles, and my values. It's not a magic fix. But it means the seed at least has sunlight to grow toward.

Of course, training AI on yourself is only as good as the self you're training it on. If you're feeding it the performing, achievement-seeking, approval-hunting version of you, you'll get polished emptiness back. Which brings us to a different kind of problem.

Mistake 2: You're Destroying Your Meta-Thinking and Learning

At my old gamification design job, I experienced one of the most hilarious uses of AI I've ever seen.

I was video calling a colleague about how they saved our last client meeting. The meeting had been tense: the leader on the client side felt our deliverables were clashing with their vision. The air had that specific corporate thickness where everyone is being polite but nobody is actually listening. Then my colleague pulled out the perfect thing to say. And just like that, the tension evaporated like steak off an American's dinner plate.

"How the hell did you do that," I asked.

"Oh yeah," my colleague said, grinning. "I went to all the transcripts from our last two months of meetings. Then I told AI the situation and asked it what I should say."

I belly laughed. Deep, cackling laughter. "Oh, that's a good one," I said, wiping a tear from my eye. "So what did you really do?"

My colleague smiled at me deviously.

I looked at them, jaw hanging. "Oh."

Even though AI saved the day in that meeting, I couldn't believe they used it that way. Because here's the thing: the client management skill that was needed in that moment, the ability to read a room, name the real tension, and say the thing that unfreezes everyone, that's a framing skill.

And if you outsource framing, you never build it.

The research world calls this cognitive offloading: using AI to replace your need to think rather than enhance it. The core issue is that framing the problem is a huge part of solving the problem. The output is not where the value lives. The value lives in the thirty minutes you spent staring at the ceiling figuring out what you actually needed to ask.

So what do I mean by framing? A frame is the lens through which you sense, perceive, feel, think, act, and relate to reality. Imagine a frame as the glasses you wear in any context. Meta-thinking is taking off the glasses and looking at them. But most of the time we aren't aware of our glasses even being there. Kind of like how you don't notice your pants against your tailbone until someone mentions it. You're welcome.

Your relationship to emotions is a frame. What friendship means to you is a frame. Evolutionary theory is a frame. You need framing skills in your classes, your relationships, your career planning, your self-understanding. Everywhere.

The framing skill doesn't just disappear when you close your laptop.

You get in the internal habit of not framing. You stop building meta-thinking skills in relationships, in career decisions, in how you understand yourself. The rust spreads. A student who lets AI frame every essay gradually loses the ability to frame a difficult conversation with their roommate. The muscle is the same, and it atrophies the same way (check out my article on how AI is rusting your college freshman brain to learn what to do about this).

As your meta-thinking and learning atrophies your framing ability atrophies. And in effect your ability to give AI the right frames for great output atrophies. Which leads us to the next mistake.

Mistake 3: You Don't Treat It Like a Person

After Claude Opus came out, I was sitting with my friend Juan at The Chesterton House talking about AI.

"You know, your articles are so much more human than other people's articles," he said, leaning back in his chair. "That joke about the golden retriever. It's all those AIs infecting everyone's writing. But when I read yours, it's just so you."

I took a sip of my coffee and smiled guiltily. "Thanks! But you should be thanking Claude. We've written the last thirty articles together."

Juan's face did this beautiful slow crumble from admiration to betrayal. Most students' writing doesn't pass what I now call the Juan Test, and the reason isn't that they're using AI. It's that they treat AI as a vending machine. Insert prompt, receive output, submit. They think: it's trained on all of human knowledge, so if I just feed it my class notes and a decent prompt, I can call it a day.

No. Please god no.

My AI writing sounds like me (sorry to toot my own horn) because I treat the AI like a person. A really, really intelligent person. And just like with people, the quality of what you get back depends on the quality of what you put in.

We think because AI is trained on all of human knowledge, more knowledge fed into the prompt means better output. The opposite is true. AI, like humans, must separate relevant information from irrelevant information. Extra context isn't just useless; it's actively damaging.

Humans are remarkably good at this parsing because it's tied to our survival: when a branch snaps in the forest, you don't process every leaf on every tree before deciding whether to run (unless you're a poetry major). But AI isn't coded to survive and thank goodness for that because I'd be the first ChatGPT target in an AI apocalypse. AI needs help to parse relevance from irrelevance.

And that's where you come in.

Your job isn't to give AI boundless information; It's to curate relevance.

What does that look like in practice? I run different AI setups for different tasks, each with context tailored to that domain.

My writing AI is trained on my voice, past articles, and a document of specific things to avoid. My information finder AI is trained on my Truth Compass framework. My gamification AI knows the Octalysis Framework inside and out. Each of these works because the context is narrow, opinionated, and relevant to the task.

I'm not going to tell you how to context engineer well because other articles will do a better job. The point here is the mindset, not the method.

The mindset holds something quietly profound. The best AI users are people with strong relational intelligence: they can read tone, adjust communication style, repair misunderstandings, and iterate collaboratively. These are the same skills that make someone a good partner, friend, or therapist. So if you want a romantic partner just get really good at prompting your chatbot (this is a joke, mostly).

This relational intelligence leads directly into the last and most currently relevant mistake.

Mistake 4: You Don't Use AI as a Team

I have a confession. I despise ChatGPT.

For years I was an advocate of Claude over the worshiped OpenAI model. Where ChatGPT acted like an ADHD golden retriever, Claude acted like a hyperintelligent thinking partner.

One morning I was evangelizing my love for Claude to a friend when she told me to try prompting ChatGPT to generate an image. "It's gonna be bloody shit," I said, "just like everything else it does." Intrigued despite myself, I asked it to make a thumbnail for Conscious College in the style of Violet Evergarden.

As you can see, it's bloody fantastic. Would have made Vincent Van Gogh blush.

I sat there awed, both at the thumbnail and my own ignorance regarding how much the models had advanced while I was busy being loyal to one of them.

I learned something important that day: stop thinking about AI as a single tool and start thinking about it as a team.

Every AI model is different. They're trained on different data, optimized for different tasks, and carry different strengths and weaknesses. Claude writes circles around ChatGPT, but ChatGPT's image generation is in another league. Perplexity is better at web search than either. This isn't permanent, the landscape shifts constantly, which is exactly the point.

The future of AI use looks less like one student with one chatbot and more like one student managing a small team of specialized agents. Yay, we all get to be managers! At least now the tears of our exploitation are water in data centers instead of actual human labor. I look forward to the day technology and consciousness are mutually beneficial.

I already do this with my writing workflow: one Claude instance trained as a writer produces the rough draft, a second instance trained as an editor reviews it, and then I incorporate both into the final version myself. The human stays in the loop not as the executor but as the designer, the one deciding what quality looks like and which output actually serves the goal.

Your value in this arrangement is less in development and more in design.

The judgment of what's good, what's relevant, what's true: that's the human part, and it's the part that can't be offloaded. So become a scientist about it. Test multiple AIs on the same task. Compare outputs side by side. Figure out which model you want as your primary for each type of work. And don't be afraid to feed the output of one into another for feedback. Sometimes the best use of a second AI is to bash the first AI's output to smitherines.

It Was Never About the Output (Conclusion)

If there's one thing I hope is clearer from this article, it's that the output was never the problem.

Most students treat bad AI output as a tool issue. I just need to prompt it better. But that's a small sliver of what's actually going on. The root challenge is you.

Building your self-understanding so you can be the sunlight that gives the AI seed direction. Protecting your meta-thinking skills so you control the frame AI operates within rather than letting it frame for you. Developing your relational intelligence so you can treat AI as a collaborator, not a vending machine. Growing your capacity as a manager so you can orchestrate a team of models instead of worshiping one.

It was never about the AI output. It was always about building your own consciousness.

If you found this post interesting you would love my free College Freshman Cosmic Journaling Kit (CJK). ✨📚

It's a gamified journaling system that helps you grow your emotional intelligence, self-understanding, and purpose with over 1,000+ journaling questions in just 15 minutes a day.