The Question I Didn’t Know I Had
I never fared well with the school system. Not because I lacked curiosity — if anything, I had too much of it. I always wanted the deep version of things, not the pass-the-exam version. The problem was I had no way to get there.
The traditional path for that kind of depth is a PhD: years under a mentor who steers you, who knows which question to ask before you even know you need to ask it. I didn’t take that path. Instead, for years, my way of chasing knowledge was the same as most people’s — open a browser, search, read, repeat.
That works, up to a point. But it has a blind spot nobody talks about: you can only search for the questions you already know you have. A search engine is fantastic at answering “what is X.” It’s useless at telling you “by the way, you’re missing Y, and you’ll need it to understand X properly.” Without a mentor, without a structured curriculum, you don’t know what you don’t know. You end up with knowledge full of holes, and worse, you don’t even see the holes — because nobody’s there to point at them.
I remember hitting exactly this wall with quantum mechanics. There were a few conceptual points that never clicked. At the time, I did what most self-learners do: I accepted it, filed it under “things I’ll get eventually,” and moved on. In hindsight, it wasn’t a missing fact I needed — it was a shift in mindset, a different way of framing the problem, the kind of thing a good mentor gives you in one sentence and a stack of textbooks can’t.
This is where AI changed things for me, and not in the way most people talk about it. I don’t mainly use AI to get answers to problems I already know how to state. I use it to build the structure I never had. I can ask for a study plan on a topic, have it organized into a real curriculum, and — this is the important part — actually chase down a chain of questions. Not one question and done, but “wait, that implies something I don’t understand, explain that too,” five or six times deep, until the mindset shift actually happens. That quantum mechanics sticking point I’d shelved years ago? That’s the kind of thing this back-and-forth finally resolves.
Three things about it specifically replace what a mentor gives you, in a way a search engine never could.
The first is patience with the questions you’d never ask out loud. There’s an embarrassment tax to asking “the basic thing” in front of a professor or in a room full of people who seem to already get it. That tax is exactly what keeps a real hole in your understanding hidden for years. With AI, you can ask the same concept five different ways, badly, repeatedly, until one framing finally clicks — with zero social cost. I didn’t get through that quantum mechanics block by asking one clean question. I got through it by asking a clumsy version, then another, then another, until the right one landed.
The second is that it follows your actual confusion instead of a syllabus. A textbook’s order is fixed — chapter 3 before chapter 4, whether or not that’s where your gap actually is. Your confusion isn’t linear, it branches sideways into things the syllabus never planned for. Chasing a chain of questions five or six layers deep only works because nothing forces you back to the “correct” chapter order. You follow the thread exactly as far as it goes.
The third, and maybe the most mentor-like of the three, is that it can diagnose why you’re stuck instead of just telling you what you’re missing. That quantum mechanics gap wasn’t a missing fact — no amount of re-reading the textbook paragraph would have fixed it. It was a missing frame, a different way of thinking about the problem. A good mentor spots that difference immediately; a search engine can’t, because it only answers the question you knew how to phrase. Describing my confusion honestly to AI, instead of just asking for the definition again, is what actually got me the reframe I needed.
It also goes beyond talking. I’ve used AI to get hands-on with things I’m learning — not just conceptual explanation, but actually building and testing, which is how I learn best anyway. That combination — structure plus depth plus the ability to actually do the thing — is what a mentor gives a PhD student. I never had that. Now I effectively do.
I want to be clear about what this isn’t. I’m not using AI to write my RTL for me, or to hand me a finished design so I can skip the work. The whole point of building a pipelined RISC-V CPU from scratch as I move from analog IC design into digital was to actually learn digital design, not to have something that looks like I learned it — AI acted as the mentor asking “did you consider this?”, not the ghostwriter doing the assignment, and I closed timing on real hardware with my own understanding intact. The same approach is running in parallel on a DSP mastery track, where AI generates the problems and I derive the math and the fixed-point RTL myself, and on a from-scratch SDR board project where I’m writing my own RTL instead of dropping in premade IP I don’t understand. Different domains, same method: AI builds the structure and asks the next question, I do the actual work.
If you’ve ever felt like autodidact learning tops out because you don’t know which questions to chase next, this might be the actual unlock. Not a shortcut to the answer — a way back to the kind of structured, deep, mentored learning that used to require a very specific and privileged path to get.