Training a GAN on 3,000 Personal Photos to Explore AI's Reflection of Memory
Merzmensch · x · 2026-08-13
The author shares their project #MERZmory, conducted in 2021-22, which involved training a visual generative model (GAN) on 3,000 personal photos.
The goal was not to have the AI reconstruct or derive visual memories (since photos are merely captured fragments of experience), but rather to observe what the AI would do with them. The author reflects that human memories are inaccurate shadows that never truly record events, only reflect them, and the most authentic memories remain such by being untouched.
Related event: Developer Trains GAN with Personal Photos to Reconstruct Memories(2 posts)→
More from Fun
- Joke: Rename Data Centers to Supercomputers to Fix Image Problem — tekbog · 2026-08-14
- Agent Arcade: Transparent desktop overlay games to play while waiting for AI agents — DanWahlin · 2026-08-14
- Live stream: Using a tool in unsupported ways to see AI infer intent — dfinke · 2026-08-13
- Grok Iteration Test: Version 4.6 Crushes Previous Build in Minecraft Spatial Judgment — adonis_singh · 2026-08-13
- Waymo Reaches Milestone with First Public Ride Directly to a User's Home — reed · 2026-08-13
- The Dilemma of AI Memory: Should Models Hide the Liquor Store? — TheZvi · 2026-08-13