Researchers at Open AI have taught artificial intelligence how to play Minecraft using only walkthrough videos. The training method was based on preliminary video training (Video PreTraining, VPT), ​​which made it possible to use a large amount of unlabeled data for this. To control AI, I used mouse and keyboard emulators.

At the first stage, the neural networks showed 2,000 hours of marked-up video with Minecraft gameplay. Here, the researchers demonstrated which buttons users pressed during the game. This allowed the system to learn to guess which buttons are pressed. At the second stage, the neural network has already viewed 70 thousand hours of unlabeled video (without data on the buttons being pressed). All entries were taken from open sources.

As a result, the neural network was able to learn not only to walk in the game world, but also to extract resources and create objects from them, run, swim, avoid obstacles, hunt animals and get food. AI has also learned to jump up and put blocks under the character to climb higher.

In the third stage, the researchers asked the users who took part in the project to help them. Players have created a new world in Minecraft, collected the necessary resources to start and crafted the essentials. All this was recorded on video, which was then shown by neural networks. Some have even built basic shelters, and the neural network has mastered this skill. The researchers used reinforcement learning, which allowed the AI ​​to eventually create a diamond pickaxe.

Scientists believe that the method of preliminary video training will allow you to quickly train neural networks for the necessary tasks, and will also allow you to train artificial intelligence to use the mouse and keyboard. The source codes of the project have already been published.

Previously, it was reported that Minecraft was completely completed in less than 2 minutes.

Lu Weibing, president of Xiaomi Group China and general manager of the Redmi brand, previously confirmed that the Redmi Note series will be updated every six months. At the same time, once a year, the focus will be on performance and new features.

In the first half of this year, Redmi launched the Note 11T Pro series, which was performance oriented. The Redmi Note 12 series, which will be released in the second half of this year, is positioned as a “small flagship” and will focus on new experiences.

A well-known insider under the nickname Digital Chat Station has revealed that the Redmi Note 12 series smartphones will use a flat screen with a hole for the front camera in the central upper part of the display.

The Redmi Note 11T Pro series uses an LCD screen, while the Redmi Note 11 Pro series uses OLED screens. In both cases, FullHD + resolution and a high image refresh rate are supported.

In addition, Digital Chat Station reports that Redmi Note 12 will receive a significantly improved image processing system. The Redmi Note 11 Pro+ uses a 108-megapixel Samsung ISOCELL HM2 image sensor with a 1/1.52-inch optical format. Probably, the new smartphone will receive an enlarged image sensor.

Digital Chat Station was the first to accurately announce the specifications and release dates of Redmi K50 and Xiaomi 12.

According to LeaksApplePro, Apple is preparing three new iPad Pro tablets equipped with the Apple M2 single-chip system, which outperforms the Apple M1 by 18 and 35% in CPU and GPU speeds, respectively.

The most interesting of the trio will be a tablet with a screen diagonal of 14.1 inches, which will be released next year. The sixth-generation iPad Pro comes in 11-inch and 12.9-inch screen sizes this year. In addition to the Apple M2, they will also get a 5G modem and a Mini LED screen. The tablets will start at $800 and $1,100 for the 128GB versions.

As for the 14.1-inch version, it also features a Mini LED screen, a minimum storage capacity of 256 GB, and a price starting at $1,400.

Presumably, new tablets will receive a screen with a notch-type cutout, which iPhone smartphones will begin to abandon this year.