We strive to inform readers not only about events in the Bitcoin industry but also to report on what is happening in adjacent technological fields — cybersecurity and the world of artificial intelligence (AI).
ForkLog AI has gathered the most important AI news from the past week.
- Media: within a year Yandex will exhaust the supply of semiconductors required for its servers.
- Chinese researchers have laid out a roadmap for AI development in the \”Era of Large Models\”.
- Elon Musk has contemplated creating his own “free” social network.
- Nvidia unveiled an algorithm to turn 2D photos into 3D images.
- Boston Dynamics released the Stretch warehouse robot for general sale.
- South Korea will test hydrogen-powered autonomous buses in urban settings.
- The most important AI deals of the past week.
Media: Yandex will exhaust its supply of semiconductors within a year
Because of American sanctions, Yandex may exhaust the stocks of semiconductors and other components necessary to run its servers within a year.
According to Bloomberg, the tech giant has faced a growing shortage of equipment due to U.S. import restrictions tied to Russia’s invasion of Ukraine. Sanctions on dual-use technologies, used in both military and civilian spheres, have inflicted the most significant damage to the drone development division, the publication writes.
The shortage of server equipment could also affect other areas of Yandex’s business, including its search service, cloud business, music streaming, and the e-commerce segment.
Additionally, the tech giant has faced a “brain drain,” as Russian employees seek to move abroad amid worsening economic isolation. Since the war in Ukraine, up to 70,000 IT professionals have left Russia, and in April this figure could rise to 170,000.
Chinese researchers developed a roadmap for AI
One hundred Chinese researchers developed a roadmap for AI development in the \”Era of Large Models.\” It runs to 200 pages and includes a bibliography of 1,637 references.
The roadmap identifies three main problems facing modern AI systems that require solutions:
- Lack of model transferability. Since the algorithm is built and trained for a specific task, it cannot be applied to solve another problem, so training the neural network from scratch is required.
- A lot of manual work. Training a model, including parameter tuning, requires the involvement of a large number of AI engineers.
- Shortage of high-quality data. This can limit applications of algorithms in some areas.
The document also discusses addressing \”the main risk of such algorithms\” — all their defects propagate to subsequent neural networks and could eventually spread to the entire community of foundational models.
Elon Musk expressed a desire to create his own social network
The American businessman Elon Musk “seriously considered” creating a social network with open-source AI algorithms and a commitment to freedom of speech.
A Twitter user wrote a post suggesting that the Tesla CEO develop a new platform. Musk responded that he is serious about it.
Am giving serious thought to this
— Elon Musk (@elonmusk) March 27, 2022
Earlier, the billionaire asked users whether the algorithm should be open source. More than 82% of nearly 1.2 million participants answered positively.
Twitter algorithm should be open source
— Elon Musk (@elonmusk) March 24, 2022
Musk also asked his followers whether they believe the social network upholds free speech principles. More than 70% of the roughly 2 million participants in the poll answered negatively.
Yandex fired American testers of autonomous vehicles
The company Yandex laid off more than 20 American employees involved in autonomous vehicle testing.
The company said that in early March 2022 the Michigan Department of Transportation paused the tech giant’s license to operate autonomous vehicles. According to The Verge, state authorities disputed this information.
Yandex confirmed that management fired six drivers who had been backing up autonomous vehicles in Michigan, and 21 employees from the team working on robot deliveries in Arizona and Ohio.
Nvidia unveiled an algorithm to turn 2D photos into 3D images
Nvidia demonstrated the Instant NeRF neural network, which turns a couple of dozen 2D photos into a volumetric 3D image.
The NeRF method was developed in 2020 by American researchers in collaboration with Google Research. It maps colors and light intensity in photos taken from different angles, and then generates data to stitch the shots together and visualize the finished 3D scene. In addition to images, the system requires information about camera angles.
The more objects that need to be transferred into three-dimensional space, the more 2D photos are required, otherwise the final result will be blurry.
According to Nvidia, Instant NeRF is the fastest solution among its peers. The model trains in a few seconds, compared to other neural networks that take several hours.
The company states that the technology can be used to train robots and drones to better understand the size and shape of real-world objects.
Boston Dynamics puts Stretch warehouse robot on sale
The Boston Dynamics company will begin selling Stretch warehouse robots to the public.
The device is a mobile platform with a seven-degree-of-freedom manipulator equipped with suction cups. The robot is fitted with cameras and sensors to navigate in space and is designed to move boxes around a warehouse and unload trucks.
According to Boston Dynamics, Stretch sold out through 2022 due to high customer demand. However, the company is now accepting orders for a new batch planned for 2023 and 2024.
Study: Facebook’s ad algorithms do not consider the context of users’ interest profiles
Researchers from North Carolina State University analyzed Facebook’s advertising algorithms.
Researchers found that models create interest profiles based on user actions, but do not account for their context. According to the researchers, if a person posts about how much they hate cheese, AI will identify only an interest in cheese and will trigger targeted ads accordingly.
To learn more about how Facebook’s interest profiles are built, researchers conducted two experiments. In the first test, they created 14 new accounts, controlled demographics, and tracked the generated list of preferences. The researchers found that 33.22% of assumed interests were inaccurate or irrelevant.
For the second experiment, researchers recruited 146 participants from around the world. They collected data from each account and analyzed the AI-generated list of preferences. As a result, 29.3% of interests did not match real user preferences.
Also, most respondents were not even aware of Facebook’s Ad Preferences Manager, the researchers said.
AI helped Google Maps remove over 100 million edits on maps
In 2021, Google Maps, with the aid of machine learning, blocked over 100 million abusive edits on maps.
According to the company, human operators together with AI also removed about 7 million fake profiles, of which only 630,000 were based on user posts. In addition, AI prevented 12 million attempts by fraudsters to impersonate a company and 8 million cases of false claims claiming to be a business account.
Additionally, machine learning helped the Google Maps team remove almost 200 million low-quality or policy-violating photos and videos.
South Korea to begin operating hydrogen-powered autonomous buses
In April 2022, South Korea will launch a hydrogen bus with Level 3 autonomy.
The vehicle was developed by Hyundai Motor and Hyundai Mobis under government order and with government support, allocating $17.7 million over two years.
According to Business Korea, from 2017 to 2021 Korean authorities invested $119 million to create nine key components for active driver-assist systems. This led to the registration of 120 patents related to cameras, radars, and lidars.
By 2027, the South Korean government plans to roll out Level 4 autonomous systems.
Researchers used AI to improve weather forecasting
Researchers linked data from smartphones with satellites to improve weather forecasting.
Camaliot, funded by the European Space Agency, developed an Android app that, via GPS receivers, collects information on the time and location of satellites. The system notes the time taken to obtain signals and uses this data to determine the smartphone’s location.
Researchers suggested that in this way they could extract more information about the atmosphere, including how the amount of water vapour affects the propagation of satellite signals through the air.
The week’s most important AI deals
From March 27 to April 2, 2022, AI startups attracted over $253 million. Here are the most interesting deals.
- AR glasses maker Nreal raised $60 million.
- Yami raised $50 million to expand its online Asian marketplace for goods and groceries.
- Pinecone raised $28 million to develop a specialized database for data scientists.
- Kaiyo raised $36 million to support its online marketplace for second-hand furniture.
- Churpy, a startup automating accounts receivable, raised $1 million.
- Sourceful raised $20 million to grow a platform for supply-chain transparency.
- Zepto raised $18.8 million to scale its instant payments platform.
- Skan, developers of software for automated business-process planning, raised $40 million.
Also on ForkLog:
- ForkLog AI launched a TikTok channel.
- Researchers found more than 1,000 LinkedIn deepfake profiles.
- Clearview AI released an updated facial recognition system.
- Google will apply AI to detect personal crises through search queries.
Subscribe to ForkLog’s AI news on Telegram: ForkLog AI — all AI news!