Google, NASA launch AI model to map methane emissions from space
The MAPL-EMIT model, which the tech giant and space agency introduced in a scientific journal, was found to detect 50% more methane leaks than human experts.
A Balatro enthusiast claims to have trained a Google fruit fly brain simulation to play the game, achieving a success rate of 20% according to a shared video. The fruit fly brain, mapped by Google, has been used for various purposes, including turning the fly into a day trader and teaching it parallel parking. ActualAerie1011, the Reddit user who shared the video, states they developed a trainer algorithm to discover seeds that can potentially lead to high scores in the randomized Roguelike game.
The brain's success rate is currently at 20% on a random seed at White Stack/Red Deck difficulty. The user claims the model only sees what's on-screen and that training is ongoing. However, some commenters have cast doubt on the project due to the lack of detailed information about the training process or references to other open-source projects.
Balatro's complex interactions between jokers and their scoring, particularly at higher difficulties, make it challenging to train a model to consistently play the game.
Written by urgent.news from Tom's Hardware's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
