April 24, 2025
Blockchain

Tech giants stood up to deep fraudsters, Zoom learned to recognize gestures and other news from the world of artificial intelligence

  • April 24, 2022
  • 0

We strive not only to inform readers about events in the bitcoin industry, but also to talk about what is happening in the relevant technological fields – the

We strive not only to inform readers about events in the bitcoin industry, but also to talk about what is happening in the relevant technological fields – the world of cybersecurity and artificial intelligence.

ForkLog AI gathered the most important news from the world of AI last week.

  • GitHub has blocked Russian developers’ accounts.
  • The tech giants have formed an alliance to fight the deep crooks.
  • Google Lens has learned to recognize text in the PC version of Chrome.
  • Instagram algorithms are trained to “put more value” on original content.
  • DeepMind introduced a new language model with 70 billion parameters.
  • They will build an underground city with its own data center in South Korea.
  • Top AI deals from the past week.

GitHub starts blocking Russian developers’ accounts

GitHub IT project storage service has started blocking accounts of Russian developers and companies.

According to media reports, the platform is suspending the accounts of programmers who currently or previously worked at organizations that fell under US sanctions without warning, effective April 13, 2022.

As of April 15, GitHub has blocked dozens of Russian accounts, including Sberbank AI Lab, Sberbank, Alfa-Bank and individual developers.

“When a corporate account is blocked, access to warehouse data can be protected for two hours, while a personal account is instantly limited when suspended,” Habr said.

Tech giants form alliance against deep fraudsters

The C2PA alliance of IT companies will develop a standard for verifying the authenticity of images and videos on the Internet. Consortium members include Adobe, Microsoft, Intel, Twitter, Sony, Nikon, BBC and ARM.

According to the Alliance, the implementation of an open standard will prevent the spread of deep frauds. Adobe says many of its authentication mechanisms will appear in 2022, and within a few years an ecosystem will be created to verify the authenticity of photos and videos at all stages of their lifecycle.

The technical standard will ensure that content is not tampered with using distributed ledger technology. Adobe says the alliance has a long way to go before all platforms can implement effective content verification mechanisms.

TSMC will launch 2nm chips in 2026

In 2026, TSMC will launch the first chips based on 2nm process technology.

The company’s CEO, Xi Xi Wei, said that N2’s manufacturing nodes will rely on insulated gate transistors (GAA). However, he did not provide any details about the architecture or technology name.

It is also known that chip fabrication will be based on existing ultraviolet photolithography with a numerical aperture of 0.33.

TSMC plans to develop test chips by the end of 2024 and begin high-volume production by the end of 2025. This means that the company’s customers will receive the first N2-based chips in 2026.

Meta Quest learned to recognize gestures

The Meta Quest headset has been taught to recognize complex gestures such as “high five” and clapping.

The company has updated its mixed reality hand tracking system to support more gestures. The new version of the Presence Platform API will become publicly available after preliminary testing by several developers. The technology also promises more complex and reliable interactions on the Meta Quest headset.

The Presence Platform update aims to improve motion tracking by using the headset’s cameras instead of the default hardware controllers. The system focuses on recognizing gestures when part of the user’s hand is out of sight of the tracker, including clapping, high five, and crossed arms.

According to The Verge, in the future, the company plans to improve the identification of a removed thumb, pinch, grab, and poke.

Google Lens adds OCR to Chrome for PC

Image search service Google Lens has been taught to work with text in images in the PC version of Chrome.

The user needs to hover over the image, right-click, click “Find with Google Lens” and select an action in the command bar that opens.

Google Lens is in the desktop version of Chrome. Data: Google.

The tool can highlight, copy, voice and translate text on images. It also allows you to find the source of the image.

This feature is available in Chrome 100 for Windows, macOS, and Chrome OS users.

Instagram is enabling original content promotion algorithms

Instagram Algorithms educated “more value” original content.

According to social network head Adam Mosseri, writers who create something from scratch should get more recognition than users who get posts from other platforms like TikTok. Developers are tweaking their content sorting tools to highlight original photos and videos.

Mosseri said the company has already done everything possible for the algorithms to “guess” the authenticity of the content, but there is no guarantee that the decisions the models make are correct.

Zoom adds gesture recognition to send emojis

Motion recognition has been added to the desktop version of Zoom.

To send the corresponding reaction to the interlocutor in the video conference, it is enough for the user to raise his hand or thumb.

This feature is available in version 5.10.3 of the Zoom app on Mac and Windows. The option is disabled by default, but anyone can enable it for any user, account, or entire group.

DeepMind introduced a new language model with 70 billion parameters

British artificial intelligence lab DeepMind presented a predictive optimal Chinchilla language model with 70 billion parameters. According to the company, the algorithm outperforms the GPT-3, Gopher, Jurassic-1 and Megatron-Turing NLG in a wide range of tasks.

The developers said that Chinchilla has the same computational budget as the 280 billion parameter Gopher language model DeepMind previously created, but has four times more data. The algorithm requires significantly less computation for fine-tuning and inference, which greatly simplifies further use, they added.

According to MMLU test results, Chinchilla achieves an average accuracy of 67.5% on prediction tasks, which is 7% higher than Gopher.

Performance comparison of different language models. Data: DeepMind.

An underwater city with its own data center will emerge in South Korea

South Korean scientists have proposed to build an experimental underwater city with its own data center. The project, worth 30 million dollars, was designed for 5 years and 80% of the construction costs will be covered by the state.

Underwater city concept. Data: KIOST.

The implementation of the underwater settlement is being carried out by the Korea Institute of Oceanology and Technology (KIOST), with the support of the Ministry of Maritime and Fisheries, as well as 23 other agencies and private companies. The facility is planned to be located near the urban agglomeration of Ulsan.

The project includes the development of technologies and equipment for the analysis of the marine environment, the construction using drones, the creation of underwater energy sources and communications, the monitoring of the health and safety of the inhabitants of the city.

The data center will help develop the underwater energy efficient and sustainable data centers of the future. Researchers will also test the system for early warning of natural disasters and use of the complex for underwater tourism. In the future, the creation of a separate hotel is considered.

Engineer created a drone interceptor of other drones

A Lithuanian engineer has developed a drone with a Drone Interceptor network to intercept other drones.

The system is a quadcopter with an increased power-to-weight ratio. The device can overtake and scout a suspicious drone, transmitting the video in real time to the ground operator.

If a potentially dangerous object must be dropped, the Drone Interceptor is placed below the target and, at the pilot’s command, separates all four propellers from the frame. As the propellers move upwards, they reveal a Kevlar mesh that entwines and hits the “enemy” drone among themselves.

Left unscrewed, the main part of the quadrocopter descends on a parachute and emits an audio signal to simplify operator search.

The most important AI deals of the past week

From April 17 to April 23, 2022, AI startups exceeded $550 million. Here are the most interesting trades.

  • Jarvis ML received $16 million to develop a platform that helps brands customize products, services and experiences.
  • The creators of the Convoy market, which connects cargo shippers with drivers, raised $260 million.
  • AlphaMountain.ai has raised $2.7 million to develop threat intelligence for cybersecurity platforms.
  • Mutiny has raised $50 million to develop a marketing service to personalize websites.
  • Prezent.ai received $20 million to expand the presentation tool’s development team.
  • Laiye has raised $160 million to develop a platform to automate office tasks of varying degrees of complexity.
  • Unlearn.AI, creators of digital twins, raised $50 million for clinical trials.

Also on ForkLog:

Source: Fork Log

Leave a Reply

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

Exit mobile version