I was an above average student in school. Scored well in most subjects. But social science? Couldn't stand it. My problem was the dates. 1857, 1947, 1789. I'd stare at these numbers thinking, "How does memorizing this help me with anything?" Nobody gave me a good answer.
Turns out, the answer was: it doesn't. Not anymore.
How the game kept changing
When I was growing up, if you didn't remember something, you didn't know it. Your brain was the only hard drive you had. Then the internet made facts free, and the skill shifted from memorizing to making sense of what you found. Now LLMs have shifted it again. Tools like Claude and ChatGPT have already read, organized, and connected the information. The synthesis is done before you even show up.
So what's actually left for us to be good at?
One question, two very different answers
I was confused about L1 and L2 cache memory, the kind inside your CPU. I typed "What is L1/L2 memory?" into Claude. Got back a wall of jargon. "Set-associative mapping." "Cache coherence protocol." I understood zero percent of it. Tried ChatGPT, Gemini. Same technical dump.
Then I reframed: "Can you explain L1 and L2 memory with an analogy?"
Claude compared it to a chef in a kitchen. L1 cache is the small counter next to the stove, tiny, but you grab ingredients instantly. L2 is the pantry down the hall, more storage, takes a few extra seconds. Main memory? The grocery store across town.
Instantly clear. Same AI. Same knowledge. Different question, completely different outcome.
Here's another one. A friend was trying to learn about investing. He asked an LLM, "Explain mutual funds." Got a textbook paragraph he glazed over. Then he asked, "I have 50,000 rupees I won't need for 5 years. What are my options and what are the tradeoffs?" Now the answer was specific, practical, and actually useful to him.
The AI didn't get smarter between those two questions. He did.
This works at work too
I write code for a living, and I see the same pattern there every day. A colleague once spent an entire afternoon trying to fix a slow program. He kept asking the AI, "Why is my code slow?" and getting vague tips he'd already tried.
I sat with him for five minutes, looked at what was actually happening, and asked Claude something much more specific: "I have a spreadsheet with 2 million rows and I'm processing it one row at a time. What's the fastest way to process all the rows at once instead?" Got a working solution in thirty seconds.
Same tool, same problem. His question could've been about any slow program ever written, so the answer was about any slow program ever written. My question told the AI exactly what was wrong, so it could tell me exactly how to fix it.
The bottleneck moved
We've gone through three eras of learning in a single generation. Memorize facts, because they were scarce. Synthesize information, because the internet made them abundant. Ask the right question, because AI has already done the synthesis.
An LLM can hold more information than you'll read in a lifetime. But it just sits there until you ask. A vague question gets a generic answer. A sharp question, one with context and constraints and the right framing, gets you something you couldn't have found on your own.
And social science?
I hated it as a kid. Dates, places, events, all abstract and lifeless.
Then I went to Sri Lanka. I walked through the places where the Ramayana is believed to have happened. Ashok Vatika, where Sita was held. The temple at Chilaw, linked to Ravana. I'd read these stories as a kid without caring much. But standing there, in the actual places, something shifted. The characters stopped being textbook names. The history stopped being a list of dates. I wanted to know more, not because someone told me to, but because I was there.
Travel gave me what school didn't: a reason to be curious. And once I had that, the right questions followed on their own.
That's really all I'm saying here. Learn to ask better questions. Our ancestors did the same thing. We just get to do it at scale.
Comments (0)