ImagineThat

Powered by AI Logic

Writes the proposals. This is the model doing the imagining. It produces one strategy per round, and successive rounds travel further from the obvious answer. Larger models tend to hold coherence longer as the ideas get stranger.
Labels and rates each proposal. It names the core approach in a short phrase, then rates how far the idea departed: conventional, unconventional, impossible, or incoherent. It never discards anything — every round is shown regardless of rating. An impossible idea is not a failure; only incoherent is.
5
How many rounds to run. Later rounds are generally more imaginative.
conventional — already on the expert's list unconventional — coherent, not standard impossible — absurd, but names a mechanism incoherent — vocabulary where a mechanism should be
Prompt
Weight-space view of every answer