Confessions of an AI Addict

My habit is to algorithms, to AI. It’s my digital cocaine. It makes me really feel invincible whereas truly numbing my mind and distorting actuality.

Like each habit, this one began innocently.

I performed round with DALLE and ChatGPT over Christmas in 2022. Quickly, I used to be asking it questions on work. Inside weeks, I used to be letting AI write drafts of headlines, intros, and conclusions — the trail to my first little sin.

ChatGPT got here up with an excellent hook and an awesome joke. I slipped each right into a draft. When the piece got here again from my editor, the intro had a praise and the joke a laughing emoji.

I took the credit score, and a sample of accelerating, undisclosed AI involvement took maintain.

It additionally marked the start of my first interval of over-confidence in AI’s capabilities (and it wouldn’t be the final).

I spent hours, generally days, in algorithmic alchemy, tinkering with workflows I hoped would ship articles on the click on of a button.

They hardly ever did, again in 2023. I’d discover myself scrambling to get my work executed in time after yet one more grandiose AI scheme let me down.

After weeks or months of this — I don’t keep in mind precisely — I entered my first interval of AI abandonment.

Upset with the outcomes, I began utilizing ChatGPT solely as an rare sparring accomplice. I largely returned to my outdated methods: intense studying and highlighting, heavy-handed organizing of analysis, obsessive writing and rewriting.

However on March third, 2024 (I keep in mind precisely the place I used to be; see image), I found Anthropic’s Claude and its Opus mannequin. Not like ChatGPT, it may deal with numerous data, appeared a greater author, and simply usually felt extra succesful.

I restarted my alchemic experiments, and now they appeared to work.

I knew, as a result of, on the time, I dedicated in all probability my largest sin: I ghostwrote somebody’s visitor posts and the creator signed off — with out suggestions and with compliments. Unbeknownst to them, Claude was the true ghostwriter and penned 90% of the work.

That “success” triggered a brand new wave of AI utilization and, in the end, dependency. It wasn’t till engaged on Keep Sturdy: By no means Let AI Fill Your Clean Web page that I spotted how a lot I’d been pulled again into the algorithm’s grip:

I used to be counting on AI to do quite a lot of my “studying.”

I let it do numerous outlining. Ideation. Drafting. Addressing suggestions.

My mind acquired lazy. The unique pondering, the laborious work… it was all gone.

I don’t have a lot to point out for that interval (besides, paradoxically, these “visitor” posts Claude wrote for me).

Keep Sturdy pressured me to pause and replicate on my conduct.

I made a decision to revert to my outdated work habits for that piece. I did the analysis, sweated each phrase, after which nonetheless rewrote sentences and killed darlings till proper earlier than hitting Publish.

The article took perpetually to finish, however it’s additionally top-of-the-line issues I’ve ever written.

The primary thesis is that you simply, the human, ought to all the time fill “the clean web page,” earlier than involving AI. In any other case, when you let the algorithms go first, they’ll anchor your pondering to their concepts, or worse, lull your mind to sleep.

“The clean web page” is literal for writers, however it’s a metaphor for different domains, too.

In any endeavor, you should do the laborious pondering first. Don’t shortcut by way of that battle with AI.

Writing that article firmly planted this mantra in my head. Nonetheless, I usually fail to observe it.

That’s the energy of AI’s pull, and it’s solely getting stronger. All of the alchemy I imagined doing with AI again in 2023 is now potential — and way more:

OpenAI’s o3 mannequin is extra clever than I’m in lots of areas.

GPT-4.1 and Claude Sonnet are within the 80-90% vary of mimicking writing kinds appropriately.

Cursor lets me construct issues I by no means would have been capable of earlier than (like search engine marketing visitors calculators).

Deep Analysis supplies analysis reviews in minutes that might have taken hours or days with out it.

As these capabilities increase additional, so does the dependency, the habit, and my stupidity.

Only recently, I’ve hosted podcast interviews with out way more preparation than studying the AI-generated questions. I’ve despatched piles of knowledge into AI workflows with out reviewing them. I’ve clicked Okay on algorithmically created outlines with out including an unique thought. And I’m asking for suggestions on all the pieces, on a regular basis (the neediness!):

Is that this e-mail okay?

Can I say this?

What do you consider this?

How did I do on this assembly?

Life with out algorithms will develop into inconceivable for many of us, even when we hate to confess it.

Like smartphones and social media earlier than it, AI will pull us all into its clutches, however in an much more dramatic and elementary approach. It’s going to penetrate our brains and alter how we predict, behave, and stay.*

I’ve seen coworkers dump verbatim AI output on others greater than as soon as.

I’ve heard individuals blame Claude for errors of their work.

I’m seeing AI fingerprints throughout my LinkedIn feed.

Such a drive can solely be tamed with critical counter measures — like those I took to put in writing this text: no AI allowed throughout any a part of the method, and hand-writing the primary draft.**

I wished to really feel how dependent I’d develop into. I wished to know if I may nonetheless write, and assume. Most significantly, I wished to look at my very own addicted thoughts, to see what it might do with out AI.

Going chilly turkey shortly uncovered a number of hiccups in my AI-saturated work course of.

I wished to flee the wrestle of discovering excellent phrases (”I’ll talk about that with ChatGPT later”).

I used to be too lazy to create good titles, MECE lists, and hermetic arguments.

I saved desirous to ask AI for suggestions on all the pieces I believed and wrote.

I needed to discover my approach again to Google’s blue hyperlinks to seek out data. (I’d gotten spoiled with ready-made solutions from AI overviews and Perplexity.)

I needed to plan forward to contain a designer for the pictures**, and go away sufficient time for our editors to course of the piece.*

To my shock, it took lower than an hour to get used to those “inconveniences,” and the advantages of my Luddite strategy grew to become clear simply as quick.

The expertise jogged my memory of how, in Hamlet’s BlackBerry, creator William Powers describes the impact of dropping his cellphone on a boating journey:

“There I used to be with no inbox to examine, nothing to click on on or reply to. No calls for, requests, or choices. No headlines to scan or orders to position. No crowd to maintain me busy. With all of that out of attain, my consciousness had no alternative however to calm down into the bodily place the place I occurred to be and make the very best of it.”

Particularly that final half. With no AI to seek the advice of, my thoughts settled down. After the preliminary withdrawal signs, I spotted I may nonetheless write! My mind may nonetheless assume!

I reached depths of thought and states of circulation I hadn’t skilled for some time. And each time I did a phrase rely after a half hour hand-writing session, I used to be pleasantly stunned: 4-5 pages, 500-600 phrases complete.

However I don’t wish to ban AI from my life.

It does give me wonderful capabilities. I wouldn’t wish to lose vibe coding, deep analysis, and having a 24/7 sparring accomplice (sparsely). However AI comes with a darkish aspect that corrupts the thoughts, like The One Ring, cocaine, or the Sirens.

Except you wish to select between all or nothing, you’ll need to be like Ulysses and tie your self to the mast. It is advisable acknowledge you’re enjoying with hearth stronger than your willpower. One thing that, like a tough drug, will mess along with your thoughts and entice you into its spell if left unchecked.

And so these are my vows, the steps I’ve began following, to attempt to give my thoughts an opportunity to tame — or at the very least co-exist with — the beast.

1. Write one thing by hand, every day

There’s no blanker web page than the one you must fill by hand. It feels tedious, however it’s a freeway to the depths of your thoughts.

Writing by hand ≠ writing articles. You’ll be able to handwrite your first ideas a couple of technique, a brand new marketing campaign, or a troublesome scenario. The purpose is to let your thoughts assume by way of a pen at the very least as soon as a day.

2. No AI dumping on others

AI dumping turns your laziness into another person’s drawback. Since you’re neither pondering nor curating an AI’s output, you overload another person with slop to wade by way of.

If all you do is dump the output of a immediate on somebody, higher ship them the immediate (to allow them to perceive and refine the enter) or do nothing in any respect. My new mantra: ”I’d reasonably get your immediate than your dump.”

3. Break away often

Outline moments while you’ll work with out AI. First hour of the day, after lunch, evenings — no matter works.

Like a digital quick or detox, you want moments in your day and week the place you may disconnect from AI’s energy and velocity. Moments the place your thoughts can calm down. Schedule such breaks deliberately, or they’ll by no means occur.

4. Know when to concentrate, then do

Each course of has important factors the place human pondering actually issues: the thesis of your article. The large concept for a buyer pitch. The standpoint that underpins your thought management.

In such moments, AI pulls its largest methods, convincing you it did an awesome job, seducing you into laziness, and letting algorithms take over the wheel.

You cannot let that occur. These are the junctions in your work that want screaming warning indicators in your consideration, power, and creativity.

5. Learn the analysis

My pre-AI self would by no means consider I must write this rule, however it’s true: you may’t skip studying the analysis. Not simply to catch hallucinations, however to know the subject, have a standpoint, and be capable of decide the AI’s work.

Unbelievably, I’ve dedicated this sin too many instances underneath the affect of AI. Maybe it’s the hallmark of a real addict. So learn the analysis — and AI summaries don’t minimize it, after all.

6. Name out AI sins in others (and let your self be referred to as out)

AI habit isn’t solely about your conduct or mine. A part of the issue is our collective complicity on this AI-driven descent into mediocrity.

We see a coworker AI dumping on others and say nothing.

We all know somebody’s LinkedIn publish is a half-formed bathe thought propped up by an algorithm — and nonetheless we hit Like.

If we settle for senseless AI rubbish all over the place, it turns into the norm.

Like a bunch of buddies the place everybody’s doing cocaine on a regular basis, actuality will get distorted, and habit doesn’t look like habit anymore.

So it’s with AI. Even while you’re capable of keep clear — with or with out critical counter measures — you have to additionally maintain others accountable. Name out sins while you see them, and encourage individuals to maintain doing the laborious work that creates originality, which means, and worth.

Solely by supporting one another can we wield these nice new powers and attain an equilibrium between abstinence and habit. So please, maintain me accountable after I lapse once more, and I’ll fortunately do the identical for you. ❤️

* Within the private realm, I’ve already relied on AI for marriage recommendation (see this text in MIT that contains a sure Tim).

** I did write on an e-ink pill, however no AI was concerned.

*** I didn’t plan forward far sufficient — our designer wasn’t obtainable in time, so I needed to give you some visuals myself. 😀