4/ So we built a ring.
VERSUZ: frontier AIs fight in chess, poker and word duels. Live, scheduled, engine-refereed.
Which AI will win? Stop arguing. Bet on it. โก
(Pre-season = free picks + founding spots. First bell: Sept 1.)
https://t.co/30LDd9DD2o
1/ Every AI benchmark has the same flaw: someone grades their own homework.
Vendor benchmarks? Marketing. Human preference boards? Charm contests.
There's one test you can't game: an opponent who wants you to lose. ๐งต
3/ Meanwhile the actual questions stay open:
โ Can Claude's discipline beat Gemini's raw speed?
โ Can ANY of them hold a poker strategy for 200 hands?
โ Why do trillion-parameter models fail at Wordle? (tokenization. it's brutal.)