The Optimization Paradox

You can always improve something, even if it's by 1–2%.

I started realizing I have this problem not when I start any project, assignment, or even this blog post-but when I finish it.

I realize that I can make changes, and then even more changes, and then add a whole new section. I never get fully satisfied, and I start contemplating whether it's perfect enough. I start noticing small mistakes that most people won't bother reading. I start overthinking the whole design.

Trying to optimize everything is annoying not only because you spend time making changes that don't really make the overall product better, but because you then start looking for feedback-especially from AI.

The AI Feedback Loop

We usually never had access to constant feedback. Whenever I needed someone to read my paper, I would ask my mom or a friend. I needed someone else to find my mistakes or just make whatever product I created better overall. However, now we can ask AI for feedback, and that's a little bit daunting.

AI sometimes edits code, ideas, and writing to make it more presentable-but it can also lose your voice.

I can give a great example:

This year, I've been focusing on building a brand for myself, which is why this is on my personal website. Building a brand for me includes blog posts, podcasts, GitHub, and even LinkedIn posts. I've been posting frequently on LinkedIn-around once every week-and it's been going great. However, LinkedIn is known for AI slop.

Posts are often just a person coming up with an idea, and then AI writing it-and it's so easily detectable. AI tends to write posts with similar formats, the same cliché, cringy endings, and it adds things most people never say.

The Experiment

I used AI to optimize my posts. I would write drafts on my laptop and then paste them into GPT or Claude, and it would point out all the flaws. Then I would start to agree with it because it knows "best"-but does it really?

I made a conscious decision to stop relying on AI so much and just write posts from scratch, and only use AI to fix grammar-without rewriting anything or even giving me feedback. I did that, and usually my posts get around 1,000–5,000 impressions, which is pretty solid.

The post I wrote without AI got roughly 30,000 impressions in one day.

Now, obviously there are so many factors with LinkedIn, and I understand that post could've just been more viral, or it could've been traction and luck. But even if that's the case, I think instead of trying to optimize the perfect post-down to ideas and structure-I just trusted my CS knowledge and wrote something likable that resonated with a lot of people.

The Takeaway

I think the main takeaway is that AI constantly gives us feedback. It's programmed to satisfy us, but also point out flaws. When we fix the flaws, we lose our own voice, and it damages our products.