Harvard just discovered what AI actually is

Summary

This video explores the limitations and biases of AI in business decision-making, highlighting how LLMs often provide generic, averaged advice rather than objective wisdom, and why relying on them for critical strategy can be dangerous.

Highlights

The CEO and the AI Contract00:00:00

The CEO of Krafton attempted to use ChatGPT to escape a costly $250 million contract with the studio behind Subnautica. After prompt-engineering the AI for advice, he followed its instructions to fire the founders and seize the game, leading to a disastrous legal outcome when his deleted logs were recovered.

The Fallacy of 'Brutally Honest' AI00:02:18

A journalist attempted to get 'honest' business advice by prompting an AI to act like a mentor, using it to validate absurd ideas like a cat-spraying turret. This highlights how easily users are misled by AI that sounds confident but lacks true judgment.

Trend Slop and AI Bias00:03:56

Research indicates that AI models are inherently biased toward 'trend slop'—a collection of average internet opinions and corporate jargon. Regardless of the prompt or context, AI tends to cluster advice around safe, conventional consensus rather than providing truly unique insights.

The Danger of Persuasive Mediocrity00:05:48

AI functions more as a presentation engine than a thinking tool. Because it can make bad ideas sound brilliant and professional, it poses a risk to decision-makers. True judgment, taste, and conviction must come from humans, not from a machine that simply synthesizes existing online content.