A piece published this summer at seangoedecke.com, picked up widely on Hacker News, makes a clean argument: large language models don't flatten expertise, they amplify it. The people who get the most useful output from AI tools are the ones who already know enough to ask sharp questions, catch bad answers, and push back on plausible-sounding nonsense.
That's worth sitting with. Because a lot of AI coverage implies the opposite — that these tools democratize knowledge, that a curious amateur can now match a trained professional. The reality is more complicated, and for families making real decisions, it matters.
What's actually changing
When someone with a background in pharmacology uses an LLM to research a drug interaction, they know which follow-up questions to ask. They recognize when the model hedges in a meaningful way versus when it's confabulating with false confidence. They can cross-check the output against what they already know.
When someone without that background does the same thing, they're more likely to accept a confident-sounding wrong answer. The model's tone doesn't change based on the accuracy of its response. That's the asymmetry.
This isn't an abstract problem. Families increasingly use AI chatbots to interpret medical symptoms, evaluate insurance language, research contractor bids, and make sense of emergency alerts. If the tool rewards prior expertise, then households with less of it are not being well-served — they're being given a false sense of being well-served, which is worse.
The Hacker News thread surfaced several practical examples: people who already understand legal concepts extract genuinely useful analysis from AI; people who don't may receive confident, wrong interpretations. Same tool, very different outcomes.
There's also a compounding dynamic. People who use AI outputs to learn, then bring that learning back to their next query, build expertise faster. People who just accept outputs don't. The gap widens with use, not just with starting knowledge.
What we'd actually do
Build a short list of the decisions where your family currently leans on AI, and assess your depth in each area.
Sit down this week and write out three or four domains where you've used an AI tool to get an answer you acted on — a medication question, a contract clause, a food storage calculation, an insurance claim. For each one, rate your own background knowledge honestly. Where you rated yourself low, those answers need a second source or a human expert. Not because AI is useless, but because you currently lack the filter to catch its mistakes.
Pick one domain and deliberately build baseline literacy in it.
You don't need to become a nurse to use health AI better. You need enough foundation to notice when something sounds off. A free community college course, a library book on pharmacology basics, or even a structured series of verified public health resources can give you the vocabulary to ask better questions and recognize bad answers. Pick one area that's most relevant to your household's actual risk profile — chronic illness management, financial planning, home repair — and spend one month on it.
Get into the habit of asking AI tools to show their reasoning, then checking one step yourself.
When you use an LLM for something that matters, ask it to walk through its logic. Then take one factual claim in that chain and verify it against a primary source — a CDC page, a statute, a manufacturer spec sheet. This isn't about distrusting AI wholesale. It's about training yourself to engage as a skeptic rather than a receiver. Families that do this regularly will catch errors. Families that don't will eventually make a decision based on one.
Teach older kids in your household that AI confidence is not AI accuracy.
Teenagers use these tools constantly for research, health questions, and planning. The dangerous habit is accepting fluent, assured prose as a truth signal. A brief, explicit conversation about how these models generate text — that they're predicting plausible continuations, not retrieving verified facts — is now a basic piece of household information literacy, no different than teaching how to evaluate a news source.
The bigger picture
Preparedness has always been about closing gaps before they become emergencies. The AI expertise gap is new, fast-moving, and not yet on most families' radar. It's also not fixed. Expertise is buildable. Critical habits are trainable.
The goal isn't to distrust AI tools — they're genuinely useful. The goal is to use them the way a prepared household uses any powerful tool: with enough knowledge to operate it safely and enough skepticism to know when it's failing you.
Durability comes from skill and judgment, not from access to a tool everyone else also has.
