Saturday, November 26, 2022

World No 1 Tech News Website

AI has learned how to play the popular world-building game

After watching 70,000 hours of YouTube videos of Minecraft gameplay

Must read

fatima khan
fatima khan
A brand new writer in the fields, Fatima has been taken under my electric spark's RGB- rich and ensures she doesn't engage in excessive snark on the website. It's unclear what command and Conquer are; however, she can talk for hours about the odd rhythm games, hardware, product reviews, and MMOs that were popular in the 2000s. Fatima has been creating various announcements, previews, and other content while here, but particularly enjoys writing regarding Products' latest news in the market she's currently addicted to. She is likely talking to an additional blogger with her current obsession right now.

Training regimes are just one thing to consider, but having to sit through eight years of another player playing Minecraft seems pretty harsh. When the revolution arrives, OpenAI may be first up against the wall following the uprising of the robots due to the testing it’s been putting its most recent AI through to allow it was playing the regular game Minecraft.

This refers to the fact that OpenAI was the first to create DALL-E and thus is accountable for the Mini-DALL-E tribute. That’s the famous AI now forced to make memes all day long for the entire internet.

To be clear, I, for one, embrace our digital kings and haven’t kicked the DARPA giant dog in a parking area, jumped into an autonomous vehicle repeatedly, or pestered an AI assistant constantly to ask for humor larks. I’m with you and the bots.

The fact that the OpenAI neural network can create a diamond pickaxe from the back of its head is fantastic.

The lengthy blog post published on the OpenAI website (via SingularityHub) details how it got the network to play Minecraft and its exciting things. First, of course, among the seventy-nine hours worth of Minecraft gameplay videos, the company provided $160,000 for a group of contractors to develop and label 220 hours of video with labels so that the AI could comprehend what it was watching and how it was relating to its actions within the game.

The technique is known as video PreTraining (VPT), as it claims that its model can be trained on making diamond tools; according to the company, it takes a skilled human approximately 20 minutes.

image credit ign

“Additionally, the model performs other complex skills humans often do in the game,” says the OpenAI blog, “such as swimming, hunting for food from animals, and then eating the food. The model also learned the art of performing “pillar jumping,” a regular occurrence in Minecraft of elevating yourself using repeatedly jumping, and then placing a block beneath you.”

It’s also important to note that it uses the traditional interface of keyboard and mouse as well, and not a particular AI-specific version that is part of the game.

It would be complicated to build a neural network if you were viewing video clips without context. That’s why it kept contractors creating a smaller set of data in which they recorded their video and the actions they performed, including keypresses and mouse motions. The data is used to label the 2,000 hours of video and then train a model known as the Inverse Dynamics Model (IDM) with the data to be able to begin to identify the more enormous 70,000 hours of collection precisely.

These videos are the primary ingredient in training complex and diverse behaviors, as you’ll see in Minecraft. The usage of VPT is now established, and the potential of this as a learning method implies that it will be able to OpenAI declares that the process “paves the path toward allowing agents to learn to act by watching the vast numbers of videos on the internet.” It’s a question whether this is something to think about or worry about. 


More articles


Please enter your comment!
Please enter your name here

Latest article