AI might be making you dumb but it made me smart

I keep seeing articles claiming that using AI makes people intellectually lazy. Apparently if you outsource too much thinking to LLMs, eventually your brain loses the ability to reason for itself.

I’ve had the exact opposite experience. Since getting heavily into AI about 18 months ago, I am objectively smarter by almost every metric I have devised myself.

The key mistake most people make is using AI as a chatbot.

I use it as a cognitive exoskeleton.

I’ve built a personal second-brain system using GraphRAG, embeddings, agentic workflows, my emails, journals, browser history, Apple Health data and approximately 14 years of WhatsApp conversations that several people specifically asked me not to upload anywhere.

The result is effectively a digital twin of myself.

Actually, several digital twins.

I currently maintain 38 slightly different versions of me and run them against real-world scenarios before making important decisions.

For example, dating.

Using historical conversational data, attractiveness scores, socioeconomic variables, menstrual-cycle estimates, local weather and what I call the **SWin coefficient (S*x Win percentage)**, I have now simulated over 10,000 potential social encounters involving myself, my wife and my wife’s boyfriend. So far my wife’s boyfriend wins in 100% of simulations.

I initially found this concerning, but after upgrading from Llama 70B to a larger reasoning model he started winning by an even bigger margin, which suggests the original results were actually underestimating him.

Still investigating.

Could be a bug.

The important thing is that I now know this quantitatively, whereas before AI I merely had a vague feeling whenever he came downstairs wearing my dressing gown.

This is what I mean by AI making you smarter.

I no longer guess.

I know.

I’ve applied the same methodology to nutrition. Previously I would think: “I’m hungry, perhaps I’ll have a sandwich.”

Primitive.

Now I photograph the contents of my refrigerator, run a multimodal inventory agent, calculate predicted glycemic response, cross-reference my sleep score and ask three models to debate whether lunch aligns with my long-term goals.

Yesterday the system recommended: sandwich

But crucially, I arrived at the sandwich through a much more sophisticated process.

Another example: arguments with my wife.

Historically these were emotionally difficult because I had to listen to what she was saying and formulate responses in real time.