Mapping my AI mind

“We turn out to be what we behold. We form our instruments, after which our instruments form us” – Father John Culkin.

The primary time I ever used ChatGPT was December 2022, lower than a month after it first launched, making me someplace between consumer no 1 million and 57 million. I requested for 20 story concepts about productiveness focused at individuals who dwell in Lagos. I hated each single one in all these concepts, however a lot has modified since first time.

Synthetic Intelligence (AI) has been quietly shaping our lives lengthy earlier than we began speaking to it. It decides what movies we see, helps us retrieve outdated images, and predicts the climate in our pockets. As soon as, Google Translate helped me converse with a French speaker in Abidjan. However this piece isn’t concerning the behind-the-scenes programs, it’s concerning the sort of AI I converse to straight, and assume with. That is about how I take advantage of Generative AI instruments—and the way, in flip, it’s reshaping how I feel, study, and navigate each day life. 

We’ve all the time constructed instruments to increase our minds. We drew symbols to retailer reminiscence. Spoken language unfold concepts; written language preserved them. Papers. Books. Libraries. Printing presses duplicated the concepts sooner. Then got here computer systems. Then engines like google. With Generative AI, we’ve crossed into one thing totally different: instruments that remix, reply, and purpose. 

I came across The Prolonged Thoughts Thesis—the concept that if a instrument capabilities like reminiscence or reasoning, and we use it as seamlessly as we use our brains, it successfully turns into a part of the thoughts. It’s not with out its critics, however the core concept caught with me. Since Giant Language Fashions (LLMs) entered my life, I’ve examined dozens of instruments and experimented with totally different fashions. Over time, I’ve realised that my relationship with them rests on a number of important pillars. 

Language is the bridge between me and the mannequin

The extra time I spend with these fashions, the extra I realise that language isn’t simply how we discuss to generative AI, it’s typically a key to area experience. I hear oontz oontz from audio system, and a musichead hears a four-on-the-floor kick sample in a 128 bpm home observe layered with synth textures. You learn this text and see an interface, and a developer sees element timber, state administration, API calls, and frontend frameworks that maintain along with clear structure. 

Each context has its language, from Regulation to creating dinner. It’s arduous to ask the precise questions when you can’t converse the language, and even more durable to assume clearly. Working with generative AI has made this painfully obvious to me. The second I step out of my consolation zone—say, attempting to explain a tune, or perceive a chunk of city design—I really feel the boundaries of my vocabulary. 

A technique I shut the language hole is by reverse-engineering. 

After watching Celine Track’s Oscar-nominated Previous Lives, I couldn’t cease excited about the unique soundtrack Quiet Eyes by Sharon Van Etten. I requested ChatGPT to explain the tune. “A haunting, emotional ballad,” it stated. 

I pushed additional: “Okay, however in music critic converse?” ChatGPT obliged: “sonic ambiance,” “meditative,” “an emotional crescendo that lands not in catharsis, however quiet reflection.” And so forth. 

I took these phrases to a different instrument—MusicFX by Google Labs—and requested it to generate music. It didn’t replicate the observe, due to copyright restrictions, however the description helped me perceive how tone and association come collectively a bit higher. 

That complete loop—from asking ChatGPT concerning the tune, to producing one thing with MusicFX, to studying by creation—took lower than 10 minutes. This identical loop helps me assume higher in groups. I as soon as wanted to clarify a specific characteristic to a product designer; one thing I didn’t have the precise phrases for. So, I collected screenshots from totally different web sites and requested ChatGPT to assist me describe what I used to be attempting to precise to a product designer. 

Language in hand, I turned to a different instrument, Lovable, and shortly constructed a tough prototype. It was efficient sufficient for the designer to work together with and push it additional than I ever may. 

Reverse-engineering nonetheless works finest once I’m low on area language, whether or not attempting to navigate authorized jargon or product designer-speak. 

Reverse-engineering continues to work for me in contexts the place I’m low on area language and wish simply sufficient to maintain it shifting, whether or not it’s spatial structure or authorized jargon. It’s not mastery, nevertheless it’s practical fluency, and that’s typically sufficient to make significant progress. 

Creativity is simply connecting issues. Generative AI is nice at that

It’s making which means out of mismatch, taking issues that received’t all the time go collectively, and whipping them up into one thing new. The artistic thoughts shuffles between deep area experience and recent perspective. I like to consider myself as artistic, so what might be higher than a system educated on huge quantities of information and floor connections throughout disciplines and contexts? 

LLMs have turn out to be part of my artistic course of, extra as a sparring accomplice than one thing that generates completed concepts. I feed them, they usually stretch it. I include half-thoughts, and it helps me discover coherence. Right here’s a breakdown of some methods I take advantage of generative AI throughout totally different artistic capabilities.

In the end, it’s much less about offloading and extra about considering wider and testing deeper by the messiness of creating one thing new. 

I (attempt to) put my fashions on a leash

I spend a wholesome period of time learning fashions—how they’re constructed, improved, and simply as importantly, the place they fall quick. The extra I perceive their limits, the higher I can belief the instrument for what it’s good at, and keep away from blindly trusting it the place it’s weak. 

ChatGPT is my each day driver, however I’ve discovered to not depend on it each time I must confirm particular real-life occasions, whether or not it’s sports activities or monetary markets. Though its hallucinatory tendencies have improved over time, I really feel safer not trusting it solely. Perplexity, however, behaves extra like a standard search engine with an LLM layer. It cites sources and each hyperlink it sends me results in a web page I can click on by and browse myself. 

However typically, I have to work by materials in a closed setting the place I’ve to have the ability to inform the distinction between what the LLM is aware of and the data I’ve fed it. That is the place Google’s NotebookLM involves play. ChatGPT operates on all its understanding of the world, the purpose-built NotebookLM operates solely on the data you feed it. 

A leash shouldn’t be about limitation for its personal sake. It’s retaining me within the loop as a decision-maker at each course of step. 

I refine by debate

Writing is considering. After I must make sense of my ideas, I write. It exhibits me what I’m considering, the place the gaps are, and the way it all holds collectively. That’s why I strategy each first draft with scrutiny: one thing to stretch, interrogate, and refine till it’s rigorous. I deal with AI the identical method. I don’t anticipate nice outcomes on the primary strive. When individuals say their LLMs aren’t useful, I typically surprise: are they reacting to the primary immediate? 

My course of is debate. I disagree with its recommendations, and I ask it to push again. That pressure is the place the considering occurs.

I’m additionally studying that my course of is totally different once I’m attempting to doc an concept or course of, than when doing artistic writing, particularly non-fiction or journalism. When establishing a course of, my aim is readability and precision, making certain that each half connects neatly with the entire. I don’t but belief it sufficient in my artistic writing course of to collaborate as deeply, since I’m attempting to evoke or develop feeling. 

Context is my swing vote.

A decade in the past, the closest you would go to seeing a mirrored image of your self on the Web was taking a “What Form Of Bread Are You” quiz, however that’s modified. A mannequin begins out generic, however slowly, you deliver it into your world. It accumulates context immediate by immediate till at some point, you’ll ask it, “based mostly on all the things you realize about me, inform me x” and it’ll offer you an insightful reply, typically with references. 

For this reason ChatGPT stays my each day driver. I check new fashions each different week, however discover my method again to ChatGPT each day. It is aware of sufficient of my spouse’s meals preferences that I solely must ship a photograph of an awesome menu for it to make an correct takeout advice. It will probably make a superb guess of most of what’s in my fridge, as a result of it’s seen the within of it. If I can’t determine between two shirt matches, it might choose the one with a greater silhouette based mostly on my physique kind. 

In a typically unnerving method, it has turn out to be a residing reminiscence, a refined however highly effective presence, an prolonged thoughts. The creepiest manifestation of this got here a number of weeks in the past, once I requested ChatGPT what Ogboni was. Ogboni is a standard socio-political and non secular fraternity amongst Yoruba individuals. ChatGPT responded intimately, together with the titles in Ogboni. I knew that already, however I do that with ChatGPT; ask it one thing I do know the reply to, to see if it will get it appropriately. 

“Based mostly on all the things you realize about me,” I requested, “what would my title be if I have been one of many Ogboni?” It stated I’d be Apena. It was a cool reply. It was additionally my great-great-grandfather’s title—I’ve by no means shared this details about my household with ChatGPT. 

In December 2022, ChatGPT was a generic responder—in April 2025, it’s turn out to be a private collaborator. 

The ambient layer

For now, my relationship with my LLMs requires effort. I select to achieve for it. I open a plain doc after which chat on a break up display screen. I provoke the dialog. It’s not going to be like this for lengthy. 

Lots of the know-how that has turn out to be a utility tended to start out out. We went to cybercafés to make use of the web; now our equipment are linked. You went to a particular web site or web page to make funds on-line, now you simply pay at checkout. 

On Google Docs, the place this piece was written, Gemini’s four-pointed star is ready to assist, whereas Grammarly is catching my spelling errors with out asking. In my neighbourhood, I’ll typically be seen taking walks whereas speaking into my cellphone and making up ridiculous eventualities with ChatGPT, as a result of a transcribing characteristic makes low-friction, high-information-capture attainable. 

The subsequent stage is when it will get proactive.

Some instruments are already higher at proactivity than others. Grammarly, the AI grammar instrument, is already chasing the errors on each textual content field in my browser. However at some point, it doesn’t really feel very far. My Fitbit will have the ability to inform my journaling app what my exercise ranges are. Maybe my journal will have the ability to generate insights like, “You are inclined to have higher days together with your temper while you cross 8,000 steps the day earlier than and had a superb night time’s relaxation.” 

And issues will shift even additional. My AI stops being one thing I take advantage of, and begins turning into one thing that strikes with me. It learns my rhythms, observes my patterns, and helps me make sense of them. The extra seamless it will get, the much less I’ll discover it.

1000’s of years in the past, it was pen to paper. Then got here recorders. Then got here computer systems. Now, it appears like all of them without delay. And this time, it listens, learns, and most significantly, thinks with you.