I don't want to be perfect, AI
Why we shouldn't use AI to replace our human-ness.
Welcome back to Jab’s Lab, my weekly musings on all things tech, startups, business, and life. Today, I’m talking about AI…again. It’s just too interesting not to. This time, however, I’m not talking about software, or everything that it can do for you, but rather what it shouldn’t do for you. In a world where AI may be better than any of us at most things, there is still a tremendous amount of value in still showing up, however imperfectly. Let’s dive in.
Here’s a question: when you have a perfectly ubiquitous tool to do any reading, writing, or thinking for you, what value do you bring?
I’ve been grappling with that question frequently this week as I have come to the realization that Claude 3.7 is a better coder than I am. It’s a humbling, existential, and difficult realization that many software engineers are dealing with right now.
AI is fantastic at taking existing patterns and thoughts and using them to synthesize seemingly new code, writing, or art. Claude is a better coder than me not because it is particularly creative, but because it has been trained on all public code to ever exist on GitHub1. I guess I learned to code in much the same way, but with a smaller sample of reference code. So of course Claude is better than me.
But just because AI can do something better than me doesn’t mean that my unique perspective is not valuable. I believe the real value that we bring as humans, whether it be code, or writing, or art, comes from our imperfections.
AI is perfect.
We are not.
Whereas we struggle through a creative process, AI takes an input, and has been perfectly trained on all of the available information on the public internet how to respond. It is the perfect amalgamation of all of humanity’s collective knowledge, and yet, it often feels so stale. Its perfection makes it a bit boring. Where is the flavor? Where is the style? In reducing all knowledge to a handful of models, we’ve managed to remove the beautiful idiosyncrasies that make us interesting.
It’s often the subtle skews on something, the small imperfections that make it real, authentic, and fun. No human is perfect. No communication is perfect.
And yet, I see a temptation to replace ourselves with a perfect AI output.
The bad habit I’ve formed is to go to Claude2 with a loose idea, prompt it in a lazy way, and get something close to what I want as the output. There is an anchoring effect when AI writes something before I’ve done all of my thinking on a topic. While I had complex thoughts and ideas, perhaps I read the AI output and think “that’s close enough to what I wanted to say, and it is written much more eloquently than I would have written it.” So I go with that instead of my own words.
But what I’ve realized is that the thought in my head was so much more vivid and real, it had a life of its own, and the writing Claude does for me doesn’t quite hit the same notes I hoped it did. Something is lost in translation.
In the world where typing “write me a nice, engaging email to this person to accomplish X, Y, and Z,” of course it’s tempting to do so. However, it feels disingenuous, and often leaves me yearning for more. It’s replacing my own thoughts, ideas, opinions, and creativity with someone else’s. And while the output email may sound better, it is not me.
I have found that through my own weekly writing, I have honed my thoughts a bit more effectively. However, these thoughts and ideas are still loose strands of connected bits, and connecting them is hard. It’s much easier to try to get AI to connect those strands, but it never seems to do so quite right.
I believe there is value in imperfect writing. And sharing these imperfect pieces of art is vulnerable. But at the end of the day, it still feels better to create imperfectly than to not create at all.
So send the email without a full AI re-write. Write a silly message and hit “send.” Film the video and put it on your story. It’s not perfect, and that’s why it’s so fun. It’s okay not to be the perfect, pristine version of yourself, especially in this world of AI-generated content. Just be yourself.
Let’s all continue to be imperfect. Vulnerable. Human. Because otherwise, every interaction we have will look like this:3
Until next week,
-Cory
And as the developers of Claude realized a majority of their output tokens are dedicated to code, they decided to optimize it even further. Don’t worry, “output token” is just a fancy way of saying “the text that AI generates.”
I am a huge Claude advocate, it is immeasurably better than ChatGPT in my opinion. If you haven’t tried it, please do and report back. I’d love to hear your thoughts.