Mini-Lecture Explainer

How to understand what AI does well and know what it gets wrong

Your tournament felt democratic. AI uses something that looks similar but works very differently. Here's the comparison.

The Prompt Tournament

How your class picked the best prompt

VS
How AI Picks Its Answer

How a language model picks what to say next

✓ Similar

Louder isn't always better

In your tournament, confident pitches won early rounds. In AI, frequently repeated patterns in training data score highest, even if they're wrong. Popularity ≠ correctness in both cases.

✓ Similar

Past patterns drive decisions

Your class built on whichever ideas won earlier rounds. AI builds each word on what came before it in the conversation. Both systems are shaped by what already happened.

≠ Different

AI has no opinion, at all

You actually believed your idea was better. AI doesn't believe anything. It has no preferences, no curiosity, and no stake in whether its answer is useful or true.

✓ Similar

Recommendations aren't neutral

When AI recommends a product, song, or answer, it's reflecting what was most common or most clicked in its training data. Just like how the loudest room shaped your vote.

≠ Different

You can change your mind

After the Socratic, you might realize a quieter idea was actually better. AI can't reflect on its own answers. It doesn't know when it was wrong unless you tell it.

≠ Different

You have values. AI doesn't.

Your group cared about fairness, creativity, or what would actually stump the AI. AI doesn't care about any of that. It just generates what statistically fits.

💡

The big idea to take away

AI sounds confident for the same reason the winning idea in your tournament sounded right, because it was repeated, reinforced, and presented without hesitation. That doesn't make it true. The difference is that you can notice when you were wrong and update. AI just keeps going.

Use these to drive the Socratic debrief
? Was the tournament a fair way to find the best prompt? What would have been fairer?
? If AI picks answers based on what's most common, what kinds of ideas might it never give you?
? When you ask AI to recommend something, a song, a restaurant, a product, who or what is it actually reflecting?
? Is a confident answer from AI more trustworthy than an uncertain one? Why or why not?
? What would it mean for AI to actually have an opinion? Does it matter that it doesn't?
? How is the way AI "decides" similar to how social media algorithms decide what you see next?