Why ChatGPT is getting worse
AUTOU
OUT 03, 2025
2025 is flying fast.
At the beginning of the year, AI still felt like a "cool tool." By mid-year, it was the only thing that matters. Hype, fear, and opportunity.
In July, I went with AutoU to VivaTech in Paris, one of the biggest tech events in the world.
Everywhere I looked: AI, AI, AI.
That's when it hit me, there's no going back.
I've been in tech for 10+ years, and I've seen a lot. But this time felt different. The demand for AI was (still is) exploding. Companies everywhere wanted it.
At the same time, OpenAI was hyping the release of GPT-5. Everyone believed it would be a turning point...
The Reality Check
When GPT-5 finally came out, it was supposed to be a revolution. Instead, it feels worse.
At best, it is a slight improvement over GPT-4. As a heavy user, I actually feel it got lazier.
The reason? Costs.
Running these models is insanely expensive. The more compute you spend, the better the answer. But OpenAI is a for-profit company, and ChatGPT is a mass-market product.
Most people use it free or pay $20/month and use it for simple questions — so they optimized the new model for cheaper, faster responses.
In other words: they made it "lazy" on purpose.
I get it. It's business. But it's also a wake-up call for us. The pace of AI progress won't always be straight up.
No Shortcuts
This is where we hit the bitter truth. There are no shortcuts.
As humans, we want the magic pill, the quick way, the easy solution. And AI companies know this, they package their products like magic.
But great results still require effort.
AI is like a calculator in math class. Yes, it gives you leverage and superpowers, but you still have to understand the formulas and do the work.
AI is the same. It's a powerful ally, but it won't build results for you on its own. You still need to understand your processes, map them, and implement the right solutions.
And the truth is, most companies struggle to do this alone.
MIT recent study found that bringing in external partners for AI projects can double the chances of success, simply because someone with experience can keep you grounded, avoid common traps, and move faster.
The Technical Ceiling
At the end of the day, what we call "AI" today is mostly Large Language Models (LLMs). They are statistical systems trained on massive amounts of text to predict the next word.
That's it.
It looks like magic, but it's not. They don't "know" anything, they just generate text.
Which means there's a ceiling. More data, more compute, better tuning — but still just text generators.
This doesn't make AI any less transformative. It's huge. But we need to demystify it, so we know where it helps, and where it doesn't.
So, What Now?
AI is not magic. It's a tool. One of the most powerful tools humanity has ever had.
Like electricity, the internet, or mobile phones, it changes the game.
But the rules of reality still apply: you need skill, you need work, you need to build.
And when you combine effort with AI, things that were impossible before become possible today. Processes can be automated. Ideas can scale. Results can compound.
That's where the real opportunity lies.
Final Note
I wish there were magic pills. Life would be easier. But that's not the case.
We must keep learning, improving, and putting in the work —> while using AI as leverage.
And honestly? That's the fun part.
AI may not be magic. But it's here, it's powerful, and it's opening massive opportunities.
See you on the journey.
PS: If you want to see one real, useful application of AI at scale, check our case for AI application with Nestlé.
The AI Playbook is where I share hard-earned insights on AI, business, and growth. Subscribe free to stay ahead.