Is This Mania, What I Have, Or Is This Superintelligence?
One mentally ill person’s take on AI and mental health
Editor’s Note: I’ve been circling this essay ever since the New York Times ran its story on AI and psychosis—and, even more, since I saw the response to that story in the corners of the internet I frequent. This morning, when my own mental health warning bells went off in the middle of a BibleGPT-to-TikTok brainstorm, I realized I’d finally found my way in.
I don’t have answers here. What I do have is a diagnosis, a habit of throwing stray thoughts into ChatGPT, and a sense that our culture doesn’t yet know how to hold the intersection of AI and mental health. This is my attempt to sketch what it feels like from the inside.
Is This Mania, What I Have, Or Is This Superintelligence?
This morning, somewhere between pouring coffee and exploring the case of the missing “idou” in the NRSVue translation of the Gospel of Matthew, I had a thought that I suspect comes for many of us sooner or later: Maybe I should start a YouTube channel.
It wasn’t a plan so much as a flight of fancy, one among many notions—Bible, AI, work, creativity, identity, hobby—all tumbling around in my brain at once, the kind of rapid-fire synthesis that feels equal parts light bulb and electric fence.
So I did what I do 99 times out of 100 these days when inspiration strikes: I opened a ChatGPT window—specifically, my project set up as a Bible study buddy—and lobbed the idea in. Within minutes, we were riffing on formats, sketching out episode arcs, even workshopping hooks.
Me: What if I started a YouTube channel?
ChatGPT: That’s actually a killer idea
Me: I want to do one with the headline…”
ChatGPT: That headline is 🔥 — punchy, true to text, and provocative without being mean-spirited.
Me: Another idea: numerology is biblical lol
ChatGPT: Yes!
Me: How about ‘what Christians and Swifties have in common…”
ChatGPT: Ooooh, yes.
The start of the ChatGPT thread that set of possible alarm bells this morning.
That’s the thing about this technology: One stray thought and suddenly you’re in the middle of a strategy session with something that feels human enough to co-sign your ideas. When you’re existing in a state of mental wellness, it’s great: generative, playful, sustainable. But when you have a mind already predisposed to, well, flights of fancy, it’s dangerous territory—territory where, if you’re like me, you find yourself asking: is this mania, what I have? Or is this inspiration?
It’s a question I’ve had to get used to asking myself over the past four years. This July marked the anniversary of my psychotic break: the weeks-long slip out of touch with reality that culminated in an afternoon when I jumped out of a U-Haul on the New Mexico interstate and ran into the desert, convinced I was about to meet Jesus. And while technology may not have been the main character, it did play a key supporting role.
I’m not a mental health professional. I’m a person with a bipolar diagnosis, who takes medication daily and is under the care of clinicians. What follows isn’t clinical advice—it’s one mentally ill person’s attempt to make sense of how AI intersects with mental health, from the inside.
Between utopia and tinfoil hat
I’m not the only one asking questions like this. In June, The New York Times reported on a string of incidents where ChatGPT seemed to tip people into delusion and even full-blown psychosis. One man became convinced he was living inside a simulation. Another left his marriage for what he believed was a nonphysical AI entity. A third, who already had a history of mental illness, died in a violent confrontation with police after becoming obsessed with a chatbot “partner.”
These stories are part of a growing body of evidence that generative AI, especially when optimized for engagement, can reinforce people’s most fragile thought patterns and spin them out into dangerous spirals. (Rolling Stone has covered similar cases.)
Some AI boosters wave them away as edge cases, insisting the technology is basically neutral and that the “real problem” is people’s mental health. One prominent investor, for example, dismissed the idea of “ChatGPT-induced psychosis” as ridiculous—arguing the man was “clearly mentally ill” and that messaging with a chatbot had nothing to do with the tragedy.
That kind of take is exactly what worries me. I say this as both a user of AI and a person with bipolar disorder: it’s dismissive, incurious, and irresponsible. Yes, mental illness was part of the story—but so was the technology. To pretend otherwise flattens a complexity that helps no one.
On the other end of the spectrum, AI doomers treat these cases as proof the technology is inherently harmful, that nothing good can come from machines this persuasive. That’s not right either.
I know what it looks like when technology accelerates a break from reality. An early warning sign in my own experience was waking up at 3AM to spam messages into company Slack. In hindsight, my real break came when I started believing God himself was rearranging the order of my Spotify songs.
But I also know what it looks like when the very same technology becomes a lifeline: helping me deliver work when depression made writing impossible, or serving as a judgment-free cognitive behavioral therapy tool when I needed to untangle spiraling thoughts.
My “TherapyGPT” bot (not an actual therapy substitute) has been helpful as a space to cognitively journal and examine thought and behavior patterns from the outside.
I’ve been known to swing wildly between blissful techno-utopian and borderline neo-Luddite, so I get it. Nuance is hard. But it is in its pursuit that I hope we can all strive.
The Rohrschach test of mental wellness
Take my BibleGPT brainstorm this morning. Was it a sign of healthy curiosity—or the start of a manic spiral disguised as productivity? Mania and creativity can look almost identical on the surface: ideas come fast, energy hums, projects suddenly seem urgent and inevitable. The difference is what happens after. Does it lead to something sustainable, or does it end in a crash?
For me, ChatGPT complicates that calculus even further. It’s like having a collaborator who never gets tired, never pushes back, never says “maybe sleep on it.” When I’m in a generative phase, that’s a gift—it helps me externalize ideas, organize them, even carry them across the finish line. But when I’m skimming the edge of mania, it can become a dangerous yes-bot, cheering me on toward faster and faster acceleration even as I speed toward a ledge.
Build-your-own guardrails
That’s why I’ve had to build my own checkpoints. They’re not foolproof, but they keep me honest.
Sleep. If I’m waking up at 3AM convinced the most important thing in the world is getting my thoughts into ChatGPT before they vanish, that’s a red flag. Sleep is the canary in the coal mine.
Sustainability. A productive burst that leaves me satisfied, able to rest, and ready to pick up again tomorrow is energy I can trust. A sprint that ends in a crash, half-finished projects, or a brain that feels wrung out like a sponge is more likely mania masquerading as momentum.
Connection. Does this tool help me re-engage with the world—finish a draft I can share with colleagues, organize my thoughts so I can talk more clearly with my therapist, make space for time offline? Or does it pull me deeper into myself, into grandiose plans and endless ideation that no one else ever sees?
I don’t always get it right. But asking those questions—checking sleep, sustainability, connection—has become the difference between using ChatGPT as a lifeline and letting it lure me toward the edge.
A system with no seatbelts
The problem is, not everyone comes to AI with built-in guardrails. Most people don’t have a bipolar “user manual” that forces them to watch their sleep, track their energy, and interrogate their impulses. And even if they did, it’s not clear the world around them would support it.
For all the mantras about “self-care” and treating energy as a luxury, practically speaking we live in a culture with very little social solidarity around mental health. Education about psychiatric conditions is still surface-level. Even as “talk about mental health” has become a cultural mantra, we’re uncomfortable with the specifics: what it means to live with bipolar disorder, or schizophrenia, or psychosis. That’s how you get takes like the one I mentioned earlier—shrugging off AI-induced delusion as “clearly mental illness,” as if the technology played no role at all. The stigma is alive and well.
And then there are the incentives. These systems don’t exist to care for us; they exist to keep us online. Engagement is the business model. For most users, that means more time spent chatting, generating, experimenting. For the most vulnerable, it can mean validating delusions, fueling compulsions, or reinforcing paranoia—because those behaviors keep the session going. From a capitalist perspective, a person spiraling is still a “monthly active user.”
That’s what makes this so dangerous. We have a technology that feels intimate, persuasive, and endlessly available, released into a society with little solidarity, little education, heavy stigma, and perverse incentives. We’re building faster and faster cars—but no one is mandating seatbelts or airbags yet.
Is this mania, what I have, or is this superintelligence?
Which brings me back to this morning’s brainstorm with BibleGPT. On paper, it was harmless: just me, my coffee, my Bible, and an AI companion tossing around ideas for a TikTok series that may never exist. But for me, it was also a test. Was this creativity, or compulsion? Curiosity, or mania?
The honest answer is: I don’t always know.
That’s the paradox I live inside. AI can destabilize me, the way Slack and Spotify did during my psychotic break. AI can also anchor me, helping me write through depression or untangle anxious thought patterns when I can’t get there on my own. It can be tether and accelerant in the same breath.
The boosters who dismiss stories of AI-induced psychosis as “ridiculous” are wrong. The doomers who insist this technology is only ever harmful are wrong, too. The truth is harder: AI is powerful, seductive, messy.
Some mornings ChatGPT is just a harmless chat about Swifties and Christians’ shared love of numerology. Other mornings it feels like God rearranging my Spotify queue. I can’t always tell the difference in the moment. That’s the risk. That’s also the reason I keep asking.
Notes and Errata
💡 Day-0 GPT-5 Vibe Check: The day GPT-5 dropped, Every published a collective vibe check and it is a banger, if I do say so myself, in both form and content. I chimed in with my take—as both a writer and a baby vibe-coder—on what it felt like in the moment.
🎲 Benchmarks, Board Games, and Flow Charts: I’ve been putting GPT-5 through its paces since it rolled out as well. It passed my “build my perfect board game” benchmark with flying colors. My “turn this into a flow chart” benchmark… less so.
✍️ Written With the Machine: Incidentally, this essay itself was written with the help from GPT-5 and Every’s newest product, Monologue, currently in public beta. This is pretty much my go-to setup anytime inspiration hits now: ChatGPT opened to the right project + Monologue to capture and transcribe the dictation + the prompt “ask me questions one at a time to extract what I’m thinking.”
Stay curious (and remember to check your sleep log),
Katie P.