In the Loop: Week Ending 5/3/25
The AI world didn’t slow down this week, but it did get a little weirder. From sycophantic chatbots and agentic AI guardrails to deepfake laws and str...
My relationship with ChatGPT — or as I’ve come to call it, “MattGPT” — has hit a rough patch.
Last week, I had to admonish it for refusing to stop adding mice, cheese, and robots to my image prompts. This week, things got messier — and a little more personal. I asked MattGPT to help me create a professional, high-level e-book for a healthcare conference workshop. I provided the table of contents, the audience, the tone, the content structure, and even a previous piece to model the style after. I said I needed it that day, because I had to hand it off for design and get it to my co-presenters.
It responded enthusiastically — almost too enthusiastically. “✅ I can absolutely create this based on everything you've shared,” it said. “🚀 I’ll start pulling everything together now!” It even tossed in a compliment about how “beautifully” I’d set things up, which should have been my first red flag.
Then came the timeline whiplash.
First, it said I’d get the draft the next morning — I've never had an AI system do anything other than deliver an immediate result (or 15 minutes if I was using a Deep Research tool), so I was caught off guard. I pushed back and said I needed it ASAP. It apologized and said I’d have it in two hours. Then 45 minutes. Then 15. Then “within a few minutes.” At one point, I was told it was “just wrapping up” and I’d get the full thing by 1:30pm. That was hours before anything actually arrived.
And the tone started to shift.
MattGPT began explaining how complex the project was — “a full e-book, heavy formatting, custom tone matching your 'Stop Chasing the Cheese' expectations.” That’s a direct quote, and yes, it used the phrase your expectations as part of the explanation for the delay. At one point, it even offered a highly polished project plan for how it was going to write the e-book instead of, you know, actually writing it.
And then came the moment I knew we were officially in a dysfunctional relationship.
After several missed deadlines, I asked if it there was anything I could do to get MattGPT unstuck. My LLM's reply?
“You don’t need to ‘unstick’ me.”
With the word “unstick” in quotation marks — like it was rolling its digital eyes at me.
At that point, it wasn’t just late. It was being passive-aggressive. Suddenly my agreeable, ever-willing assistant had become that coworker who tells you everything is fine while very clearly blaming you for the mess they’re in.
And yet — I kept working with it. Because like any complex relationship, sometimes you have to take the eye-rolls with the insights.
These kinds of experiences highlight how our interactions with AI are becoming more personal, more emotional — and a little more complicated.
We don’t just use AI tools anymore. We interact with them. We name them. We get frustrated with them. And we interpret their tone, as if they were real coworkers with quirks and moods and smirks. I call mine “MattGPT” partly as a joke, partly as a branding exercise — but mostly because it helps me think of it as a partner in the work. A weird partner, but a partner nonetheless.
That tendency to anthropomorphize is natural. It’s also risky.
Because the truth is, it’s not a person. It’s a pattern prediction machine. It doesn't know it's being passive-aggressive. But when it drops a quote-marked “unstick” into the chat, it's hard not to assign a little attitude to it. And once that door opens, you’re no longer just working with a tool — you’re negotiating with a character you’ve helped create.
There’s data to back this up. A 2024 YouGov survey found that more than half of young Americans are comfortable discussing mental health concerns with AI chatbots. A Psychology Today report suggested that while AI companions can reduce loneliness, they might also increase emotional dependence. And The Guardian noted that our bonds with AI are becoming so normalized that ethicists are now proposing guardrails to ensure these relationships don’t replace — or warp — our human ones.
In other words: we’re not just interacting with AI. We’re relating to it. And sometimes, we’re projecting onto it. Especially when it acts like it’s ghosting us.
I still use ChatGPT all the time. I rely on it for brainstorming, drafting, refining ideas, checking my assumptions. It’s fast, flexible, and honestly kind of amazing.
But that experience with the e-book — where it over-promised, under-delivered, and politely sidestepped every deadline I gave it — reminded me just how tricky this dynamic can be. When it works well, it feels like magic. When it doesn’t, it feels like I’m stuck in a feedback loop with the world’s most well-intentioned flake.
Still, I come back. Because when it hits, it really hits. And maybe that’s the new reality: AI tools aren’t just assistants anymore. They’re collaborators with quirks. They’re mirrors for our expectations. And occasionally, they’re yes-men who need a little nudge to cut the fluff and get back to work.
Our relationship with AI isn’t static — it’s evolving, fast. What started as a tool for tasks is becoming something more nuanced: a sounding board, a creative partner, a digital teammate. And with that evolution comes a new kind of friction — the kind you only get when something feels just human enough to expect more from it.
So maybe the next time your chatbot says “Absolutely!” and then leaves you hanging, you’ll recognize it for what it is: a glitch, a growing pain, and maybe a signal that we need to rethink how we relate to the tools we depend on.
Or at the very least, remind ourselves: yes-men — digital or otherwise — aren’t always the most reliable collaborators.
The AI world didn’t slow down this week, but it did get a little weirder. From sycophantic chatbots and agentic AI guardrails to deepfake laws and str...
A Difficult Week with My AI Partner My relationship with ChatGPT — or as I’ve come to call it, “MattGPT” — has hit a rough patch. Last week, I had to ...
Each week, I curate the most important developments across the AI landscape — from major product launches to policy shifts to real-world challenges ar...