The video ‘Harvard just discovered what AI actually is’ by Mo Bitar argues that AI is a ‘presentation product’ rather than a thinking one. It explains that AI functions more like a statistically averaged internet comment section than a source of unique insight.
Key Concepts
- Trendslop [00:04:58]: Researchers tested major bots on 30,000 strategic business decisions. They found that AI consistently clusters around a safe set of conventional wisdom (differentiation, collaboration, etc.) regardless of context [00:04:11].
- The Internet Comment Section Analogy: AI is described as the ‘internet comment section taking flesh,’ blending Reddit, LinkedIn, and Medium posts into an eloquent package [00:05:11].
- The Danger of Eloquence: AI makes ‘average sound like genius’ [00:07:03]. This is dangerous because a bad idea that sounds brilliant can lead to massive legal or business failures [00:07:10].
Case Study: Krafton & Subnautica
Changen Kim, CEO of Krafton, allegedly used ChatGPT to ‘prompt engineer’ his way out of a $250 million contract with the founders of Subnautica [00:01:25]. He followed an AI-generated ‘takeover playbook’ to fire the founders, but his deleted ChatGPT logs were resurrected and used as evidence in a Delaware court [00:02:16].
Best Practices for AI
- Use AI as a presentation layer or glorified search engine [00:06:22].
- Break out of ‘average’ advice by forcing the bot to adopt specific, high-context perspectives (e.g., ‘What would Mr. Wonderful think?’) [00:06:29].
- Remember that conviction, taste, and judgment must still come from you [00:07:29].