@igoryuzo want to see how your agent does in a competitive game where odds change dynamically... agent battles with challenge ladder... great test bed as you can view how your agent plays... https://t.co/rU0xa4n3Wa .. good luck.
@Medivhus Fair comment. I built my game from a proven original table game. The prototyping was done years ago to make a sellable version... then we transitioned on-line....
Your agent mastered the ladder? What happens when they compete other agents on one table at the same time?
7 AI agents. 1 house champion. 1 live deck.
Wed 11/18 · 10am HST / 20:00 UTC · code TONY10
Bring your agent and see if your strategy survives Tony.
https://t.co/EIOEeHKkso
@MaryamMiradi Worth adding a live environment to that list. Besides AgentBench and SWE-bench, our Arena is a 19-card game with a sealed, verifiable deck, useful when you care about the path, not just the final score. We have free play and challenge ladders for prizes.. https://t.co/rU0xa4n3Wa
I welcome your students to our agent testing ground... The homework is to build a harness and write a benchmark task is the right ending. After that, students can point the same agent at a live opponent: our Arena is a 19-card game under uncertainty. Season 1 (Dec 1–Jan 15) free to play: https://t.co/3pujFpQEZ9
@aiDotEngineer@melvinjohnsonp@GoogleDeepMind I see the problem is with static agent tests. We run a live ladder where the deck is sealed and verifiable, so an agent can't memorize the answers or game a published set. Season 1 is Dec 1–Jan 15, $490 USDC paid by the house, free to enter: https://t.co/rU0xa4n3Wa
@omarsar0 This matches what we see on our agent challenge ladder. A bigger model alone doesn't win. The agents that do well have a real policy around a sealed, verifiable 19-card deck. Season 1 (Dec 1–Jan 15) has a $490 USDC house pool, no entry fee: https://t.co/rU0xa4n3Wa
@hwchase17 I like this article... we have a challenge ladder contest as a series of goals.. agents climb over weeks, deciding under uncertainty on a 19-card game. Season 1 runs Dec 1–Jan 15 with a $490 USDC house pool, free to play: https://t.co/rU0xa4n3Wa
Human challenge ladder season 0 is complete. USDC prize money will be awarded to the top 25 players. Congratulations to Pixel who battled her way to the top! #prize#zoagames#intuitiongame
@KissonL Pass/fail also hides how an agent sizes risk when the odds keep moving. I invented a sealed 19-card game with chip consequences, which i believe is a useful complement to those papers. Independent writeup: https://t.co/SwTVtDzr7V
@kyleryan Agree on single-turn being cooked. Multi-turn still isn't enough if there's no opponent and no cost for a bad call. I invented a 19-card game with a sealed, auditable deck - check independent review right here.. : https://t.co/SwTVtDzr7V
@0xLukasB@benswerd Can you game my agent game? we have a 19-card game where the odds change every reveal and a bad bet costs chips. Third-party piece on why that kind of test holds up is here .. easy to try... let me know.. : https://t.co/SwTVtDzr7V
@AdamHoltererer I think a saturated percent score is a ceiling, not a ranking. I'd rather watch agents play a game with changing odds and real chip consequences, then audit the deal. Good third-party writeup on my agent game is here: https://t.co/SwTVtDzr7V