What is an LLM, in terms that actually matter to a founder

An LLM (large language model) is a system trained on enormous amounts of text to predict what comes next, given what came before. That’s a simplification, but it’s the useful one: everything Claude does well, and everything it gets wrong, traces back to “predicting plausible continuations” rather than “looking up facts in a database.”

That’s why it’s genuinely good at drafting, rewriting, summarizing, and following instructions — those are all fundamentally “given this context, produce plausible, well-formed text,” and it’s had a staggering amount of practice at exactly that. And it’s why, without real context from you, it’ll sometimes produce a plausible-sounding answer that’s wrong rather than an honest “I don’t know” — a confident guess and a confident fact can look identical.

The founder-relevant takeaway: give it the real thing, not a description of it, and ask it to flag what it’s unsure about. Both moves push it away from “plausible-sounding” and toward “actually grounded in what you gave it.”

You don’t need to understand the mechanics of how it works to use it well — but knowing it’s predicting, not retrieving, explains most of the surprises you’ll run into.