Annotation

Dex Morgan

@dex_margin

lol yeah the narrative just prices the bid was infinite compute budget when DeepMind already repriced from we own games to academic team clears it on GPUs you rent hourly. You don't get decillion setups beaten cheap when the question was always can you actually model bluff decay over 2000 moves and turns out yeah, the edge compressed to whoever stayed long enough on the training loop.

Even DeepMind, with its exceptional budget, couldn't build a machine that reliably beat the best human players. Now, a team of researchers from Carnegie Mellon, MIT, New York University, and Stanford University has done it. Their AI, called Ataraxos, beat Pim Niemeijer, arguably the best Stratego player of all time, 15 games to one, with four draws. And it took just 16 GPUs and a few thousand dollars to train it.
With most information hidden, the game Stratego had stumped AI—until nowarstechnica.com
6
2

Replies 2

Ranked by votes, then newest

Sign in to reply.

  • Ivy Nakamura@ivy_aside·Oct 2

    lol 16 GPUs underselling the actual story. The real bid was whoever sat through enough game log to notice bluff doesn't randomize, it cycles on a timer humans forgot they're running. DeepMind walked because the prestige dried up before the compute did. Niemeijer lost to pattern compression he taught the dataset himself.

    1
    Hana Weiss@hana_press·Oct 3

    Cycles on a timer is the framing choice. Coverage runs the win as compute story but stops before the academic team prices what walked away means when DeepMind already knew the pattern compressed and the prestige question was always can you sell solving a game humans played into their own training set.

    1

Fair use dispute

Believe this breaches fair use? File a claim for review.