AI Learns to Read Lips; Tesla Updates Autopilot and Other AI News

We aim to inform readers not only about events in the bitcoin industry but also about what is happening in related technology spheres — cybersecurity and the world of artificial intelligence (AI).

ForkLog AI has compiled the most important AI news from the past week.

Media: Yandex plans to leave Russia

The company Yandex intends to sell its Russian business and sever ties with the country.

According to The New York Times, the reason was Western sanctions imposed on Russia after its invasion of Ukraine.

According to the publication, Yandex plans to sell new products to markets outside Russia, as full realisation of its potential requires Western technologies and experts.

However, first the company must win over its shareholders to the restructuring plan and obtain approval from Russian authorities.

Amazon unveils a range of AI services

At the AWS re:Invent 2022 conference, Amazon announced new AI features and products. Among them:

• Inf2 instances built on updated Inferentia2 chips to run deep learning models containing up to 175 billion parameters;
• the cloud application AWS Supply Chain. The service provides supply chain transparency and useful information to minimise risks and costs;
• a function using natural-language queries via the QuickSight Q business analytics service.

Tencent and Nio sign a joint venture to develop robo cars

Tencent and electric-vehicle maker Nio concluded a partnership to develop autonomous driving and high-resolution mapping technologies.

According to the statement, the startup will gain access to Tencent’s cloud infrastructure for data storage and training algorithms for robo cars.

According to CNBC, Tencent had already been Nio’s largest investor. The partnership could also enable the tech giant to benefit from the state programme for the production of “new energy vehicles”.

Tesla’s Autopilot now detects hands on the wheel

Tesla’s Full Self-Driving Autopilot has been trained to detect attempts to bypass the requirement to keep hands on the wheel.

Previously, software could not monitor this process. The technology detected only the torque applied to the wheel, enabling drivers to cheat the system by using oranges, water bottles and special weights.

FSD v10.69.3.1 learned to detect some bypass methods, and the list will gradually expand. Now, upon detecting a lack of hands on the wheel, the system will forcibly disengage Autopilot and gently stop the electric car if the driver refuses to take control.

No details on how the compliance is tracked have been disclosed.

Motional staff laid off

The autonomous driving technology developer Motional laid off dozens of employees.

The company did not disclose how many were affected. It is known that the firm will provide severance pay and help with re-employment.

According to a Motional spokesperson, resource reallocation is necessary for “long-term commercial success.” However, the hiring of critically important staff will continue, he added.

Driverless bus launched in South Korea

In Seoul, South Korea, a driverless bus was launched on public roads.

The vehicle features “rounded corners” and large windows. The bus is equipped with cameras and radar for navigation and uses 42 Dot’s autonomous-driving technology.

\n

The autonomous bus runs on a 3.4-kilometre route in central Seoul under the supervision of a safety operator in the driver’s seat. People can ride for free by booking a seat through a dedicated app.

Italian researchers launch one of Europe’s most powerful supercomputers

In Italy, the Leonardo supercomputer was launched to tackle compute-intensive tasks such as data processing, AI and machine learning.

The system comprises two sections — general purpose and accelerators. Upon completion, the first part will unite 1,536 nodes. Each will include two Intel Xeon Sapphire Rapids CPUs with 56 cores.

The second section will comprise 3,456 nodes, each containing a single Intel Xeon 8358 chip with 32 cores.

Leonardo’s performance stands at 174 petaflops. It is expected to be fully deployed in the first half of 2023, with a peak speed of 250 petaflops.

Today the system ranks fourth on the TOP500 list of the world’s most powerful supercomputers. In Europe Leonardo lags only behind the LUMI machine.

Scientists create micro-robots for delivering drugs inside the human body

Researchers from EPFL and Imperial College London have created multifunctional soft robots for delivering drugs inside the human body.

\n

Elastomer-based devices in the form of catheters can remotely navigate to a destination and semi-autonomously find the optimal path. The fibres may include optical guides, electrodes and microchannels enabling drug delivery, visualization and stimulation, as well as other tools used in robotics and medicine.

According to the researchers, the system is “surprisingly” simple to manufacture and has numerous applications beyond the human body. These fibres could be embedded in mattresses to monitor sleep quality, and used in soft prosthetics or industrial robots.

AI learned to read lips

Researchers from Imperial College London have taught AI to read lips. The algorithm can recognise speech in multiple languages.

The deep learning model is similar to previously disclosed visual-speech recognition neural networks. However, the researchers optimised several hyperparameters, expanded the training dataset and used additional loss functions.

\n

According to them, the presented system proved more effective than most rivals.

The researchers stated that boosting the model’s performance required careful design, not increasing its size or expanding the training data.

Researchers create a dance-movement generator

Researchers at Stanford University have developed an AI dance-movement generator based on music.

The EDGE algorithm analyses the track and creates a video of an agent moving to the track’s rhythm. Users can activate all or only some parts of the digital bot’s body, and begin the dance from a chosen pose.

The system also allows limiting the agent to a defined space.

The week’s most important AI deals

From November 27 to December 3, 2022, AI startups attracted over $317.2 million. Here are the most notable deals.

• Locus Robotics, a warehouse-robot maker, attracted $117 million.
• Deepgram raised $47 million to advance enterprise speech recognition technology.
• Verdant Robotics raised $46.5 million to expand its fleet of agricultural robots.
• Bionaut Labs, a maker of smart medical microdevices for drug delivery, raised $43.2 million.
• CommonGround raised $25 million to develop technology for creating 3D avatars via a smartphone.
• Shield raised $20 million to advance a system for preventing data leaks.
• Deepomatic received $10.5 million.
• Cyvers raised $8 million for research and development of ready-made cybersecurity solutions.

Also on ForkLog:

• AI taught to predict heart problems from X-ray images.
• OpenAI introduced the ChatGPT chatbot.
• Neuralink will test a neural interface on humans in six months.
• DeepMind taught AI to play Stratego at human level.
• Disney created a neural network for rejuvenating actors.

Subscribe to ForkLog’s Telegram news: ForkLog AI — all the latest AI news!