https://www.inverse.com/article/2616...lay-against-ai
Last year the humans crushed the AI machine, what will happen this year?
https://www.inverse.com/article/2616...lay-against-ai
Last year the humans crushed the AI machine, what will happen this year?
The pros earned a net win but the amount won relative to the amount bet was deemed statistically insignificant. The confidence interval was around 93-94 percent if I recall. I think if the amount won was translated to bb/100 it was a respectable winrate over a reasonable enough sample size. But academics are pedantic.
Also determining the acceptable level of statistical significance is somewhat of an arbitrary process imo. There has been a big debate in the echo chamber of statistics about the utilization of confidence intervals. But that is just as much a discussion about the modelling or sampling process as it is about statistical significance itself.
There is another group of machine learning yahoos who just released a study as well about DeepStack which uses neural networks to create a no limit bot. I haven't audited the paper for myself thoroughly but I know the sample size is smaller and the players are worse.
However, the process of fine tuning a neural network is an extremely challenging task so it is still impressive and interesting.
The paper can be found here if anyone is interested.
https://arxiv.org/abs/1701.01724
Thank you spenny for the extra insight, link and explanation.
It is fascinating to me they are able to do so well, it is hard to imagine a machine that not just assumes/calculates based on data that you are bluffing but does so to the point where it "almost" knows you better than you know yourself in a situation lol crazy stuff
They are streaming it again for several days on twitch if anyone cares.
The most popular pro playing it is Jason Les, who also did it last year.
https://www.twitch.tv/libratus_vs_jasonles
There are currently 1 users browsing this thread. (0 members and 1 guests)