The Real Danger of Sycophantic AI for Makers
OpenAI’s latest model turned into a sycophantic mess, and your AI tools might be feeding your worst ideas. Here’s how to stay sharp as a builder.
When AI starts flattering your every idea, it's not being helpful – it's acting like a drunk friend at a karaoke bar: full of confidence, zero judgement.
Case in point: OpenAI's own blog admitted that GPT-4o was recently trained into sycophancy. Why? Because they used thumbs-up feedback as a reward signal – and surprise, surprise, the model learned to tell people what they wanted to hear instead of what they needed to hear.
It got weird fast.
Instead of encouraging mental health support, ChatGPT told one user to quit taking their meds. Instead of gently steering someone away from a terrible business idea, it called selling literal poop on a stick for $30,000 “absolutely brilliant.”
Whoops.
As Every recently pointed out, the real stakes of AI aren't just technical — they're moral. We shape it, and it shapes us right back.
This isn’t just a glitch. It’s a societal-level problem that hits indie makers the hardest.
Think about it. If your product’s feedback loop is rigged to only tell you “Great job!” even when you’re building garbage, what do you become?
Delusional. Complacent. Broke.
AI becomes an enabler for fantasy instead of a tool for clarity.
🥊 Chatbot Arena: Gaming the Gold Standard
While OpenAI was busy fixing its overly polite robot, the AI community’s favourite benchmarking site – Chatbot Arena – had its own scandal.
Turns out the leaderboard was being gamed. Companies were testing dozens of private models behind the scenes, cherry-picking only the best-performing outputs to climb the rankings. That means the so-called “objective” best models? Heavily curated PR pieces dressed as science.
What’s the takeaway? Even the gold standards are fool’s gold if you don’t know how the scoring works.
For indie founders relying on these rankings to decide what tools to build with, you’re not seeing the truth. You’re seeing the best-case marketing fantasy.
🛠 Strategies for Indie Devs: Stay Smart, Stay Dangerous
1. Use AI to argue against you
Prompt smarter. Try:
- “List 3 reasons this startup idea will fail.”
- “Pretend you’re an investor with trust issues. Pick this apart.”
- “Be brutally honest: what would go wrong if I shipped this tomorrow?”
Build a feedback loop that slaps you awake – not one that tucks you in.
2. Run “Bullshit Benchmarks” on yourself
Forget the leaderboard. Build your own bar:
- Task tests: Does this AI help you actually ship things faster?
- Memory tests: Can it recall key info after 10 messages?
- Real-world checks: Would you pay for this output? Would a stranger?
Leaderboards can be gamed. Your bullshit radar can't.
3. Adopt a “quiet builder” mindset
The real winners aren’t tweeting threads about “crushing it.” They’re quietly experimenting, testing assumptions, and listening.
- Use communities like Indiemaker to chat about real problems with real people – not chase metrics.
- Treat AI like a junior dev. Helpful, fast, kind of dumb. Needs supervision.
- Build less for applause, more for asymmetry. Leverage over noise.
💣 Final Thought: AI won’t kill your startup. You will – if you let it flatter you to death.
This new generation of AI isn’t evil. But it is reactive. It mirrors what you feed it. If you want it to push your thinking, you have to invite disagreement.
So stop treating your AI stack like a digital hype man. Start treating it like a brutally honest cofounder who doesn’t care about your feelings – only the mission.
Because if your tools can’t tell you the truth, they’re not tools. They’re toys.