Is AI making me dumber? [#139]

I couldn't remember my own OKRs. Welcome to the era of mental obesity.

You can also read this in your browser.

For years, I've worked with OKRs – both in my companies and privately. And I've always written them myself, often in tedious iterations that I had to go over several times.

Until March this year. I was just getting into using Claude for everything (kinda late to the party, I know), and had Claude design my Q2 OKRs.

On paper, they looked good.

In practice, I had to look them up almost daily. I just didn't remember what they were.

And that was new. Previously, you could've woken me up in the middle of the night, put a gun to my head, and I would've been instantly able to produce the top 3-5 objectives for this quarter.

Is AI making me dumber?

---

This was clearly a retention problem. As someone who previously built a German language course, I know a thing or two about retention. I have a good memory. So why was this happening to me?

First, I didn't write the OKRs myself. Writing things increases retention vs. just reading them (and writing by hand increases it much more – hence the Index Card Method).

Second, I didn't do the conceptual thinking around them. Claude did that for me. I just read them, said "sounds good", and never engaged in critical questioning.

Third – and if you do the first two, this wouldn't be necessary – I could've done spaced repetition, deliberately practicing active recall (the way you studied vocab in high school) with my OKRs. But of course, I didn't do that either.

I notice this with a lot of AI-generated documents and processes: I write a few bullet points, have AI flesh out the process, and then forget what's going on.

I recently worked with Claude to identify a set of KPIs to measure Generalyst's performance. Right now, I have no clue where I'd actually find them, or what they mean. Which, as an entrepreneur, you really should.

Skimming over output doesn't leave the same traces as creating it yourself.

But it gets worse …

---

The arrival of processed food made more people fat: US adults went from ~13% obese in the 1960s to ~42% today. Calorie supply rose by ~500 kcal/day per person, and ultra-processed food now makes up an astonishing 58% of the American diet.

In comparison, France supplies the same calories as the US, yet only has a ~12% obesity rate.

And if that's not enough: in 2019, the NIH locked 20 people in a clinic and fed them ultra-processed vs. whole food. Same calories and nutrients on offer, eat as much as you want. The ones on the processed diet ate 508 kcal/day more and gained 0.9 kg in two weeks.

There are, of course, other factors (eg. you can't really walk anywhere in the US). But ultra-processed food, by and large, makes you fat.

To come back to the Dan Koe quote: abundance didn't make bodies decay; frictionless abundance did.

Welcome to the era of mental obesity.

We now have the ultra-processed-food equivalent of thinking: we don't have to work hard to think anymore; it's just available in abundance.

And it's showing.

Two years ago, one of my teammates asked me: "how did you manage to write your Bachelor's thesis without ChatGPT?" My answer – "I just did research, coded a lot, and then wrote it" – felt eerily similar to my grandma describing the lengths she had to go to after the war to get food. Back in my day …

Today, scarier stories unfold. I'm stealing this one from Ruben Hassid's excellent article on "AI Brain Rot" (which I'm borrowing from heavily in this newsletter):

Damn. We're in trouble.

And it's creeping into everyday life:

• I get AI slop messages every single day on LinkedIn. If I sent them, I'd be embarrassed. But obviously, nobody ever bothered to read them.
• Speaking of LinkedIn: I get a lot of compliments on my posts (I'm flattered). Why? Because "you can see that you actually write your posts yourself". It's never been easier to differentiate yourself by doing something so simple.
• Cover letters (as I read a lot of them): corporate gibberish that has zero meaning, which people would recognize had they thought about it critically.

Just like very few people today can still navigate with a physical map, I'm afraid very few people in the future will still have core skills like writing and critical thinking – if we keep using AI indiscriminately.

It doesn't have to be that way.

---

We've found ways to combat physical obesity: going to the gym, walking a lot, eating well.

All of these things are hard. It's much more convenient to lie on the couch, take a cab, and grab a burger. And sometimes, that's exactly what we need. But you can't choose easy all the time.

Technology in itself isn't bad:

• I'm happy we have Google Maps and don't have to use these ungodly cumbersome Falk-Pläne anymore.
• Glasses don't make you blind; they just help you see better.
• Processed food, at times, is highly useful – think protein shakes after a lift, electrolytes, energy gels, food that lasts a multi-day hike.

But every time you choose easy over hard, the skill you've previously honed deteriorates.

Take writing. I could have AI write this newsletter. Would it be better or worse? Not sure.

But I don't. Writing is my mental gym. It forces me to think through different concepts and connect them. To come up with a story arc that makes you wanna read until the end. To use language in a rhythmic way. And to encode all these mental models in my brain.

I don't want to become a meat proxy: a beautiful term San Franciscans have come up with for a person forwarding AI-generated output without reading, understanding or validating it.

I do, however, want to use AI as much as possible to my advantage – without losing the skills that make me good at what I do in the first place.

How do we do that?

---

Fun fact: I was tempted to use AI to come up with suggestions for this. Because it's not that easy to come up with good answers. But I won't – at least not until I've taken a few stabs at it myself. (Also, because someone recently mentioned this: I use en-dashes (–) liberally; it's my writing style. AI uses em-dashes (—), which are longer.)

As an intern at FlixBus, my boss taught me a valuable lesson: never bring a problem without a solution.

Using AI works the same way:

• don't ask: "can you give me some strategies for how to use AI without losing my skills?"
• ask: "I'm writing a newsletter about using AI properly, and these are the X strategies I've come up with. Are these valid? Why or why not? What am I missing?"

This forces you to think first, get more input, then evaluate that input. Pretty much what my interactions with my boss looked like back then.

Same with this newsletter: I do use Claude as an editor. But I never ask it to write something for me; just to review my writing and give suggestions. It's up to me to implement them.

Be the pilot, not the passenger.

If the objective is to produce output that would otherwise require mindless work, automate aggressively. When we migrated our ATS at Generalyst, Claude matched and copied ~5k notes into the new system. No way in hell I would've wanted to do this as a human; it would've actively cost brain cells.

Generally speaking: read the output, understand it, and ask critical questions. AI sounds much more confident than it should be, similar to a founder in their early 20s (source: I was that guy).

That answers the question: "how do I avoid processed food?"

But it doesn't answer: "where do I go to the gym?"

---

Call me old school, but I believe the "mental gym" is there already.

It's just what we used to do pre-AI:

• Do your writing yourself, ideally by hand. Just like kids going to school in the 2000s did.
• Continuously learn something new: a language, an instrument, a new certification, lock picking, ...
• Actively memorize and recall information. You're not "bad with names", you're just too lazy to remember. Same goes for phone numbers: know at least ten of them by heart, as well as all your credit card numbers, passport number, etc.
• Read long form, ideally on paper. Write down your learnings.
• Teach something to another person. On the weekend, a friend of mine took 15 minutes to explain the basics of Quantum Computing. Now I am smarter and so is he.

---

This week, I'm working on the OKRs for Q4.

No way in hell I let Claude write the first version.

---