TDS JAN 2026 · PROJECT 1 ANALYSIS

What 765 Students Taught Us About Learning, Hacking, and Everything In Between

March 2026

"In January 2026, 765 students enrolled in a graduate course about AI tools. What followed was not quite an exam. It was closer to a social experiment — one that nobody designed on purpose."

The exam had twelve questions. Eleven of them measured skill. The twelfth — and this is the one nobody expected — measured whether you'd noticed that the first eleven had already been solved by someone else.

This is what 765 students, 9,046 saved submissions, and 1,731 evaluated AI images can tell you about how people learn when the answer is always one search away.

Five Stories

Social Dynamics

The Exam That Became Infrastructure

By the 154th student to solve a puzzle, all 100 secret agent IDs were already public. Only 63 students ever contributed a new one.

Contribution Strategy

Two Labor Markets

When a professor mentioned one repository in a discussion forum, 801 students redirected their contribution strategy in days. Merge probability beat public value.

Behavioral Archetypes

Six Ways to End an Exam

55 students finished the exam — and then kept touching the dashboard until the final log made them look broken.

Skill Transfer

There Were Two Exams

Among 419 complete portfolios, systems performance vs. image performance correlated at 0.020. The course had two nearly independent skill tests hiding inside one exam.

AI Tools

The Model Playbook

DALL·E was wrong for emotional charts. Gemini won on paradox. Ideogram was the dark horse for affective abstraction. The data told a story the marketing never did.

Interactive Dashboard

Image Gallery & Scores

Browse every AI-generated image from 765 students across 4 creative briefs. See Gemini's dimension-by-dimension scores, compare prompts, explore what makes a chart exhibition-worthy.

The Numbers

0 Students Enrolled
0 Saved Submissions
0 Failed Saves
0 Images Evaluated
0% Used the Public Decoder
0% Used Sink Repos Near Deadline