in

Nvidia AI plays Minecraft, wins artificial intelligence conference award

my own all the diamonds–

NeurIPS 2022 honors MineDojo for playing Minecraft when advised by composed triggers.

MineDojo's AI can perform complex tasks in Minecraft.

Enlarge/ MineDojo’s AI can carry out complicated jobs in Minecraft

Nvidia

A paper explaining MineDojo, Nvidia’s generalist AI representative that can carry out actions from composed triggers in Minecraft, won an Outstanding Datasets and Benchmarks Paper Award at the 2022 NeurIPS(Neural Information Processing Systems) conference, Nvidia exposed on Monday.

To train the MineDojo structure to play Minecraft, scientists fed it 730,000 Minecraft YouTube videos (with more than 2.2 billion words transcribed), 7,000 scraped websites from the Minecraft wiki, and 340,000 Reddit posts and 6.6 million Reddit remarks explaining Minecraft gameplay.

From this information, the scientists developed a customized transformer design called MineCLIP that associates video with particular in-game Minecraft activities. As an outcome, somebody can inform a MineDojo representative what to do in the video game utilizing top-level natural language, such as “discover a desert pyramid” or “construct a nether website and enter it,” and MineDojo will carry out the series of actions needed to make it take place in the video game.

Examples of tasks that MineDojo can perform.

Examples of jobs that MineDojo can carry out.

Nvidia

MineDojo intends to develop a versatile representative that can generalize discovered actions and use them to various habits in the video game. As Nvidia composes, “While scientists have long experienced self-governing AI representatives in video-game environments such as StarCraft, Dota, and Go, these representatives are normally experts in just a few jobs. Nvidia scientists turned to Minecraft, the world’s most popular video game, to establish a scalable training structure for a generalist representative– one that can effectively carry out a large range of open-ended jobs.”

Nvidia

The acclaimed paper, “MINEDOJO: Building Open-Ended Embodied Agents with Internet-Scale Knowledge,” debuted in June. Its authors consist of Linxi Fan of Nvidia and Guanzhi Wang, Yunfan Jiang, Ajay Mandlekar, Yuncong Yang, Haoyi Zhu, Andrew Tang, De-An Huang, Yuke Zhu, and Anima Anandkumar of different scholastic organizations.

You can see examples of MineDojo in action on its main site, and the code for MineDojo and MineCLIP is offered on GitHub.

Read More

What do you think?

Written by admin

Leave a Reply

Your email address will not be published. Required fields are marked *

GIPHY App Key not set. Please check settings

Regulative filings recommend Nvidia’s ditched RTX 4080 will return as the “4070 Ti”

Regulative filings recommend Nvidia’s ditched RTX 4080 will return as the “4070 Ti”

Patent discovers in-game “collusion” by tracking “external connections”

Patent discovers in-game “collusion” by tracking “external connections”