Tell AI about yourself and give it your data—the more it knows, the more precise the answer
37 episodes in 12 months · 52 total · first in June 2024
I constantly load huge amounts of my own data into the model: email archives, raw business data, call recordings, lab results. The more context, the more surprising and useful the conclusions: the model finds patterns I had not seen myself.
Write rules about yourself into the system—how to answer, who you are and what matters to you. And if you ask about something delicate or not about yourself, tell the backstory straight away, otherwise the model will draw wrong conclusions about you.
Happening now. The scale keeps growing: in No. 089 (December 2025) you can load more than a hundred thousand phone calls into the model as one archive, and in No. 105 (April 2026) I advise businesses not to roll out a system but to export their data and analyse it in a strong model.
Where to start
Write rules about yourself for the AI and hand it your raw data: bank statements, email, lab results.
Instead of rolling out a new system, export the company's data—calls, CRM, reports—and analyse it in a strong model.
Loading a lot does not mean the model has read it all: a large volume needs checking too.
“It gave you something. We've been over this many times, right? It gave you something. That doesn't mean it deeply, fundamentally, completely and fully studied and figured it out, and didn't miss anything, and gave it to you with precision.”
When I said it
Quotes
“I use ChatGPT basically a lot in Russian as well, and I've got a whole set of my own rules written up in it: how to answer me, how I'm to use it, what amount of text to give, and so on. It answers quite precisely, very measured.”
“…I was uploading huge arrays of raw data, in terms of analytics, some reporting from my various businesses, and asking it to work out various patterns for me. And, I've already talked about this, yeah. And the level of the conclusions is just staggering.”
“That's why practically everywhere in businesses I often say: “You'd do better not to implement a system, but to export a volume of data, analyse it with the help of, say, the Pro versions, Heavy versions, all sorts of Ultra versions, say, of Google and Grok, Claude and ChatGPT, right.””
“I mean, and you have to explain to such a degree who you are, what you want to do, what your, what your path ahead is, what you see ahead.”