I was finishing up a presentation for a tender bid. Polishing the arguments with an AI. Made sure everything was tight. Then, just for kicks, I asked it to argue the opposite. It tore my position apart and actually convinced me (the client was asking for something different, and the idea itself had some weak spots). Had to weigh the pros and cons and rework a few sections.
The conclusion I came to once again, after using AI not as a co-author but as a reviewer:
LLMs don't have opinions, but they're absolutely incredible at arguing any side of any question. And that's exactly why they're such a powerful tool for forming your own views. You have to deliberately ask the model to argue with you — not just do the whole task for you. For instance, when I'm wrapping up an important project or writing something that matters, I ask AI to evaluate the result:
- what's strong and what's weak from a logic standpoint
- where the argumentation holds up or falls apart, and why
- how a reader might perceive this style
- what could be improved structurally
It's a much better way to get useful feedback and actually learn.
The only thing to watch out for is flattery — by default, almost every model is ready to agree with whatever you put in front of it.