signalHacker News Show HN2026-09-28
Show HN: Jev-Like Model Learns to Cook
OpenJev, a decision model applying Jev's ideas, was trained on one RTX 5080 to play cooperative Overcooked. Two independently acting copies learned to serve six soups in 512 ticks using only native controls. The code and trained model are released.
- for who
- Game AI researchers and reinforcement learning practitioners
- why now
- Newly released OpenJev model learns cooperative game control from native buttons, opening fresh interaction-learning paths.
- what changes
- Enables training decision models with only native controls and natural language inference, no hand-crafted action planners
- to do
- Download the released OpenJev code and trained model to experiment with native-control training on cooperative tasks
key points
- OpenJev trained on Overcooked serves six soups in 512 ticks with two agents
- Uses natural language inference to select among six native controls
- Code and trained model are released
#decision model#OpenJev#reinforcement learning#Overcooked#open source
score
score 7 out of 10. 0-10: how dense the facts are, multiplied by how much you can do with them after reading. 8+ means the topic's evidence bar is met: benchmarks and availability for a new model, amount and investors for a funding round, revenue figures for a solo-money story. Below 5 an item does not enter the digest. A press release scores 3 or less, a reprint loses 2, anything older than 14 days loses 1, a headline that misleads loses 3.
read the source