Overview & result
Ably Labs shows Jev Pong, built on Jev: Pong where the ball moves one step per model decision. Jev against the LLMs.
From the original post
Jev Pong by @ablylabs: Pong where the ball moves one step per model decision. Jev against the LLMs.
How Jev fits in the loop
- A game of Pong in which the ball moves one step per model decision.
- Each model gets the same real state and the same question.
- A model that answers slowly makes its ball slow.
- Jev is called through Vercel AI Gateway with one AI SDK call.
How to reproduce
- Review the source documentation and repository at https://github.com/ably-labs/jev-pong.
- Configure environment with TypeSafe Jev / System One API keys.
- Run test prompts against the decision endpoints.
Why this build matters
Shows Jev in practical use for games and real-time decisions.
Limitations
- The author calls it a latency demo, not an intelligence demo.
- In the author's runs the chat models answered the same question correctly 95 to 100 percent of the time; they are slow, not wrong.
- It measures one simple question, not general capability.