The author argues that current AI chatbots excel at generating plausible-sounding statements but aren’t designed to discover or verify truth. They contrast these polished “oracle” systems with messy yet reliable collective institutions like science, warning against over-deferring to a few powerful models. Instead, they call for pluralistic, AI-augmented processes—such as community notes—to improve truth-finding without sacrificing diversity.

• AI chatbots generate plausible-sounding word sequences optimized for confidence, not verified truth, making them persuasive even when wrong
• Messy collective institutions like peer review, prediction markets, and journalism outperform polished AI oracles because disagreement and complexity help surface truth
• Even a highly accurate AI would flatten the "messy middle" of values, ethics, and culture that can't be reduced to single correct answers
• Concentrating epistemic authority in a few closed-door AI companies risks letting their commercial or ideological agendas quietly shape public belief, so pluralistic AI-augmented tools like community notes are safer than single oracles