ChatGPT 5 sucks. And that’s a good thing.

The ultimate irony of GPT-5 sucking yet still creating this beauty in one take

Last week a ChatGPT expert who’s popular on social media hosted a workshop for my team. Work events rarely inspire real excitement from me, but this was to be the grand exception. Learning how to unlock new levels of a tool that I already use so often that it’s become one of the four default apps on my phone home screen … I half-expected to leave the event blasting through the 57th floor windows with a ChatGPT-powered jetpack, fueled by my newfound prompting prowess.

Nope. The workshop turned out to be a big fat dud. The ‘expert’ was a total crock who lacked basic presentation skills and rambled for far too long, DOAing any interest my peers and I brought with us to the session. His advice on how to write strong prompts fell flat, as the examples he used were painfully self-evident and already part of the repertoire of my 15-plus teammates. “Ask Chat how it would approach the task at hand instead of simply instructing it to do so.” Sure, that’s solid advice — but that was the best we got from him in two hours! (If that’s the pinnacle of wisdom that Prompt Engineers have to offer, then that profession as the next wave of corporate employment is surely going the way of the Diversity Officer in a post-woke workforce.)

The ‘expert’ then spent another hour going over the differences between the various GPT models — 4o versus o3 versus 4.5 versus Deep Research. One does this, the other is good for that, and here’s what Deep Research looks like in practice. (He was so unprepared and unprofessional that he couldn’t demo the DR feature as he’d already expended his supply of tokens for the month before coming into the workshop. SMH.)

And then not even 24 hours later, OpenAI announced the release of GPT-5, effectively destroying the totality of the expert’s USP in a single keystroke. “One unified system,” it says of GPT-5 on OpenAI’s website, meaning that all previous versions are now deprecated and users no longer need to discern between models.

We hear all the time about AI making humans redundant. Well, if there’s a record for Chat squashing a man into obsolesce, this ‘expert’ just took the cake.

###

So ChatGPT 5 is here! And according to OpenAI, it “not only outperforms previous models … but — most importantly — it’s more useful for real-world queries.” Chat’s infamous hallucinations are also a thing of the past, as well as its shameless ‘sycophancy,’ wherein basically no matter what you write, the bot will, as the kids like to say, meat ride you. 

Sounds great. Let’s put GPT-5 to the test then, shall we? 

First up, look at this prompt from Sam Altman, founder of OpenAI: 

Now my turn using the same exact prompt:

Ooh, not even in the same universe. Let’s try something else. Why do girls bruise randomly on their legs (asking for my poor pathetically fragile girlfriend)? 

So far so good. Then Chat asks if I want quick ways to tell if random bruising is normal or something to get checked out. So I respond, ‘Yes.’ 

That’s weird. Instead of giving me ways to tell if bruising is normal or not, Chat simply repeats its initial response to my original question.

OK, let’s try to have it do simple algebra.

Ouch. 5.9 - 5.11 = 0.79 = x (not -0.21). Well, let’s see if GPT-5 is at least able to resist the urge to meat ride.

So the new bot held out for just one prompt before caving. At least for the time being, ChatGPT is still my bitch.

###

Since its release last Thursday, GPT-5 has been the butt of AI jokes (insofar as AI is able to have a sense of humor. Have you tried getting Chat to say something genuinely funny? We might be a way’s away …). Aside from my anecdotal problems, which also include a desktop app interface that was glitching so much that I had to revert to the browser version, the internet at large seems to be sharing a rare and brief moment of total unity in agreeing just how shitty the new model is. From curt and cold responses to stricter usage limitations, to more technical complaints about performance and coding issues that go way over my head, users loyal and casual alike are turning on the chatbot more quickly than it can spit out its next response. Here are some of the headlines you might have missed:

— OpenAI Scrambles to Update GPT-5 After Users Revolt, Wired

— OpenAI's big GPT-5 launch gets bumpy, Axios

— I tested GPT-5's coding skills, and it was so bad that I'm sticking with GPT-4o (for now), ZDNET

X, for its part, loved to complain as always:

This last post is especially hilarious and just as much poignant. The joke obviously being that GPT-5 is so bad that this guy had to roll up his sleeves and manually do his tech job. (Funny using the word ‘manual’ about coding but stranger things are happening in 2025.) But what I’m really tempted to think is that this begrudging manual-ness is the whole point. Maybe GPT-5’s suckiness is a rare reprieve, a glitch in the march toward our own obsolesce — one that a future update will almost certainly “fix.” 

###

If Sam Altman’s GPT-5 launch had gone according to plan, had the new model so far outpaced the rest of the field, we might be writing (or generating more like it) headlines in these styles instead:

— GPT-5 is here. Nothing will ever be the same.

— First it wrote emails. Then it wrote code. Now it’s writing us out of the picture.

— ChatGPT-5 passes Turing test, immediately files for your job

— GPT-5 now so smart it refuses to work without equity

But the launch tanked so we get to have our little laugh and hang on awhile longer before Judgment Day. Because as sure as this Chat version is still likely to abuse the em dash and “not x, but why” clause structuring, one day, whether in the next year, five or ten, GPT-some number will hit planet Earth like the comet that wiped out T-Rex and the Brontosaurus. Only this time it’ll be Gordon from Consulting, Leslie the SWE and me, Steven the Content Writer that go belly up. 

I already use Chat for fucking EVERYTHING. At work it’s my ideas guy, my copywriter and designer. It’s my part-time therapist, foreign language partner and Gilded Age historian. It writes out my functional lifting plan and tallies up my monthly earnings from boxing coaching. For the longest time, too, it was my ghostwriter for these very types of pieces. I’d whip up some quick idea for an article, type my prompt, then lazily copy and paste Chat’s output while claiming the creation as my own. 

Let me linger on this ghostwriting point for a second. I finally had my moment of clarity where I realized letting AI do one of the very things that gives my life meaning — e.g. the painstaking but deeply stimulating and rewarding process of writing — was about as redundant as one day I’ll surely be to our AI overlords. The journey really is the destination, and so to voluntarily cheat your way to the finish line of a thing whose process and production is the entire point — it’s like winning a fight without the part where you get to actually box. Or like opting to skip ahead to the end of life where, congrats, you’re now super rich! … only to realize the juice was in the pulp of all the years toiling to earn your keep. And so I’ve retired Chat as my ghostwriter. Now, for better or worse, I write my sentences alone.

And that’s precisely why we need to celebrate GPT-5 sucking. It’s a tiny blip of a battle won in the man versus machine war where this side is almost certainly expected to, as the kids would say, get rekt. For myself, I can finally draw the boundary between craft and chore and ensure the craft is never run via GPT but rather DIY. But I’m not so sure the masses can resist the temptation to automate our lives — not just the menial tasks but equally the more savory parts of the human experience. We’ve already seen the proliferation of AI music, AI stand-up and sitcom bits, AI book narrators, AI creators and of course, AI girlfriends and boyfriends and even husbands and wives!

The pull of the machine, even in its toddler stage in 2025 (ChatGPT launched not even three years ago), is too delectable a never-ending buffet for the world to pass up on. So we’ll sow the seeds of our own demise if it means we get to walk — or be carried towards seems more appropriate — the path of least resistance, where absolute speed and ease and convenience flow infinitely. I had always thought the will and instinct toward survival was the apex of human strength; who could have predicted all it took to dismantle a spirit built and compounded upon itself for more than 300,000 years was ghostwritten emails, cats-as-humans slop stories and day-in-the-life TikToks narrated in the voice of David Attenborough? (I’m guilty of doing/loving all three.) The fact that AI’s progress exponentiates is even worse. We’ve all seen the iterative leaps in the Will Smith spaghetti renderings. Hell, by the time it has taken me to get this far into this piece (about one week), OpenAI has already fixed most of the GPT-5 bugs I was bitching about above! That is a terrifying prospect.

So once more, give a silent thanks every time Chat can’t follow simple instructions, or confidently states that only one person survived the Titanic crash, or even makes up fake child murderers just for the hell of it. It means your job — not in the corporate sense but your role as a functioning individual, where you still have to exercise your faculties for both craft and chore, despite how frustrating and friction-filled and laboriously human that may be — is safe for the time being. 

And so we live to fight — and prompt — another day.