What it does
The published program runs Qwen, an AI model on local hardware. The paper reports a hosted Jev-scoring variant, but its client is missing. The default demo replays saved decisions, not fresh model calls or response times.
Benchmarks & research
Solve puzzles by keeping alternative paths and changing the available moves after feedback instead of committing to one route.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
The published program runs Qwen, an AI model on local hardware. The paper reports a hosted Jev-scoring variant, but its client is missing. The default demo replays saved decisions, not fresh model calls or response times.
Inspect a saved puzzle example in the demo to see how an action's result changes the next move or prompts another path. Reading these records needs no inference account. Fresh local Qwen runs require substantial GPU hardware; the published demo does not supply the hosted Jev client or reproduce its reported comparison.