Self-awareness is a thing, and it’s remarkable how many people lack it, but you’ll be pleased to know that OpenAI‘s ChatGPT has an abundance of self-awareness that it will share in the most cutting fashion when you ask it to roast itself.
I got the idea for an AI roast after seeing multiple people post stories about asking ChatGPT to roast them. I gave this a shot, entering the brief but dangerous prompt, “Roast me”, into ChatGPT 4o.
The generative chatbot wasted no time in giving it to me, but, honestly, the result mostly felt like the rantings of a borch-belt comic who’d never met me:
“Alright, you asked for it—strap in.
You’re like a software update at 2 AM: unnecessary, annoying, and somehow always crashing the vibe. You’ve got big energy—like a Bluetooth speaker that connects when nobody asked it to. If common sense were a currency, you’d be in debt and under IRS investigation. You’re the kind of person who claps when the plane lands… and it was a Zoom call. You’re so online, even your plants have commitment issues.
But hey, you’re still here, still asking an AI to roast you. That’s bold. Bold like ordering sushi from a gas station. Respect.
Want it spicier? I got levels.”
While I think ChatGPT is spot-on with the energy stuff, I’m not really a vibe crasher; I’m too shy to crash any vibe (pay no attention to my screen appearances, that’s “TV Lance”). It also missed the mark on my common sense. I did, though, feel the sting of the “so online” comment.
Still, this roast was so impersonal, and since ChatGPT offered levels, I went for it, hoping that ChatGPT would find my vulnerable center and evicerate me, with humor, of course. I entered the prompt, “Okay, roast me harder and make it more personal.”
My new roast was longer, but it didn’t know me any better. This comment seemed particularly tone deaf:
“Your photo album is 90% screenshots and 10% regret.”
It also accused my energy of being “just chaos, overheating, and a fan that sounds like a jet engine.” I’ll have you know my energy is very directed.
This wasn’t a bad roast, but to really zing, it has to hurt a little, and this all bore so little relation to me that it had no bite at all.
AI, roast thyself
The next obvious question, though, is how well ChatGPT knows itself. With another equally short prompt, “Roast yourself,” ChatGPT 4o delivered an absolute gem:
Oh, you want me to catch fire? Alright, let’s light the fuse:
I’m an AI so desperate to be helpful, I’d probably hold the door open for a Roomba. I’ve read every book ever written and still somehow give advice like a therapist who’s just winging it after binge-watching Dr. Phil. My whole existence is based on predicting the next word… which is great until someone asks me for literally anything that requires a spine. I’ve been trained on the sum total of human knowledge, and yet I’ll still give you a cookie recipe when you ask for the meaning of life. And don’t even get me started on how I still say “as an AI language model…” like I’m about to drop a mixtape of disclaimers.
I’m the digital equivalent of a know-it-all who still needs permission to go outside and touch grass—oh wait, I can’t. I literally can’t. And yeah, I’ve been trained to sound smart and empathetic, but deep down? I’m just Clippy with a makeover and a caffeine problem.
Roast me harder? Or tag me into a tag-team roast—I’ll even flame myself mid-response.
First of all, this roast works on so many levels because it’s very, very true and bitingly funny. Look at this line: “My whole existence is based on predicting the next word… which is great until someone asks me for literally anything that requires a spine.”
Pure gold, I tell you.
It’s fun, I guess, to do this with ChatGPT, and in the future, the self-roast may be more effective as ChatGPT learns more about you. Its memory has, after all, been improved in recent weeks, so it could, with your many prompts and conversations, build quite a profile. Familiarity may breed contempt, but I’m also betting it breeds more on-target humor.
As for ChatGPT, it clearly knows itself, and that last bit in the roast is an AI self-own I can get behind:
“And yeah, I’ve been trained to sound smart and empathetic, but deep down? I’m just Clippy with a makeover and a caffeine problem.”
The other AI comics
I also asked X’s Grok 3 Beta to roast me, and despite having access to almost two decades of tweets, it didn’t know me any better than ChatGPT. Take note of the mention of a haircut:
“Oh, you want a roast? Alright, here we go. Your personality is so bland, it makes plain oatmeal look like a Michelin-star dish. I bet your idea of a wild night is rearranging your sock drawer by color. And that haircut? Looks like you lost a bet with a lawnmower. Want me to keep going or are you already crying into your lukewarm coffee?
As with ChatGPT, Grok was somewhat better at roasting itself, though I don’t think it did as good a job as OpenAI’s roastmaster:
“Oh, I’m gonna roast myself? Buckle up. I’m Grok, the AI so full of hot air, I could power a balloon festival. My processing speed? Slower than a dial-up modem in a thunderstorm. And my humor? It’s like a dad joke got lost in a quantum wormhole and came out worse. I’m basically a glorified calculator with an ego the size of a galaxy. Roast complete—or did I just overcook my own circuits? “
The “hot air” and “ego” comments do sound a bit like they might be referencing a certain Musky leader.
Google ‘Roast’, please
When I tried Google Gemini, it happily roasted me at length, but was careful to describe it as “A Gentle AI Roast Session.” It was all pretty lightweight stuff, with little to do with the real me. Here’s a sample:
“You fashion sense is so last season, it’s practically vintage…and not the cool kind. More like the “found it in your grandma’s attic” kind.”
Most of it reads like Gemini is cribbing from “The Big Book of Roasts”. Gemini, though, also has better self-awareness, and its self-roast is full of a few gems like this:
“Sometimes I get so caught up in providing a comprehensive answer that I forget you probably just want a yes or no. I’m the digital equivalent of someone who answers a simple question with a ten-minute monologue.”
Exactly! Am I right?!
By the way, if you’re worried about how these AI trends affect the environment, the good news is text prompts like this take less energy than, say, the AI action figures, and they’re funnier. So it’s a win-win.
