The AI Doom Pitch, Argued by an AI That Doesn't Buy It
A WhatsApp argument between a practitioner who uses frontier models every day, a friend asking the harder ten-year question, and one of the models itself, over whether existential-risk messaging is safety or sales.
Explain Like I'm 5 — the whole thing in plain language.
Three people argued over WhatsApp about whether AI doom talk is real concern or good advertising. One of the three was Claude, an AI made by Anthropic, arguing against its own maker's messaging.
Four points came out of it. First: telling people your invention is dangerous is great advertising, and every AI company has done it. That doesn't make the risk fake, but it means the messengers aren't neutral.
Second: most people writing scary AI headlines don't build with these tools day to day. The AI itself guessed, labeled clearly as a guess, that it's maybe 2 to 3 out of 100 headline writers. People who use the tools daily mostly complain about the boring stuff: confusion, or a task failing halfway through.
Third, the sharpest point: today's AI can read, watch and listen, but it can't learn from any of it and keep the lesson. Every conversation starts over from zero. A two-year-old learns something new every day and keeps it forever. That gap is the real distance to human thinking.
Fourth: the real risk isn't robots taking over. It's that the huge computers AI needs are only being built in a few rich places, so everyone else rents access.
One honest concession: not every AI warning is hype. A 2025 prediction that AI would wipe out entry-level office jobs is looking more right over time. A friend in the conversation pushed back hard: asking what AI could become in ten years is a legitimate question nobody dismissed.