Slot Machine · self-consistency · JEV-Arcade
One borderline post, fifteen pulls: does the verdict hold still?. Slot Machine is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of self-consistency.
JEV-Arcade runs in the browser, so it needs JavaScript.
- JEV-Arcade · AI games: Jev vs. text models, live — Fifteen live AI games pitting Jev, a judgment model, against conventional LLMs: WikiRace across Wikipedia, chess puzzles, guardrails, function calling and more. Free and open source.
- WikiRace · AI models race across Wikipedia — Watch AI models race across Wikipedia link by link, live. Jev scores every link while text models name one; every hop, cheat and cost is shown. First to the target wins.
- WikiRace Hall of Fame — Every WikiRace on JEV-Arcade: which AI model reached the target first, in how many hops, and what it cost. Replay any race move by move.
- Legal Moves Only · typed actions — Chess puzzles where every move Jev makes is a real move. Legal Moves Only is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of typed actions.
- Switchboard · intent routing — Every caller patched to the right line, or to a human when unsure. Switchboard is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of intent routing.
- Customs · LLM guardrails — Every message through the scanner: pass, inspect or block. Customs is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of LLM guardrails.
- WikiGuessr · hierarchical classification — Where on Earth is this article? Say only as much as you're sure of. WikiGuessr is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of hierarchical classification.
- Rail Yard · model routing — Jev works the switches; the text models answer. Rail Yard is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of model routing.
- Two Truths and a Lie · verification — A text model writes the claims; Jev checks them against the source. Two Truths and a Lie is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of verification.
- Judges' Panel · composite scoring — Jev scores each quality once; you decide what matters. Judges' Panel is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of composite scoring.
- Ghost Maze · real-time decisions — Same maze, same seed, one clock. The ghosts don't wait. Ghost Maze is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of real-time decisions.
- Needle Hunt · line-by-line search — Find the line that answers the question, or say it isn't there. Needle Hunt is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of line-by-line search.
- Slot Machine · self-consistency — One borderline post, fifteen pulls: does the verdict hold still?. Slot Machine is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of self-consistency.
- Drive-Thru · function calling — Orders in plain words, out as typed function calls. Drive-Thru is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of function calling.
- Memory Match · entity alignment — Two shops, one product each: the same thing or not?. Memory Match is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of entity alignment.
- The Big Sort · map-reduce over big data — Thousands of articles sorted into topics while you watch. The Big Sort is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of map-reduce over big data.
- Blind Tasting · features for ML — Guess the critic's score: closest without going over. Blind Tasting is a free JEV-Arcade game: watch Jev, a judgment model, and text models play it live, a hands-on demo of features for ML.