Tesla autopilot crashes into plane; VK to buy Yandex media assets and other AI news

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

ForkLog AI has gathered the week’s most important AI news.

VK to buy Yandex’s media assets

Yandex will sell to VK the media assets News and Zen. The parties have already signed a contract outlining the main terms of the agreement.

Details on the price and expected timing of closing are not available. Information on how News and Zen will operate under VK’s leadership will be provided later.

To complete the deal, the companies must obtain clearance from the Federal Antimonopoly Service of Russia.

Former Lyft engineer to lead the Pentagon’s AI division

The former Lyft top executive was appointed head of the Pentagon’s AI division.

According to Breaking Defense, the former head of the ride-hailing platform’s machine-learning unit Craig Martell moved to the post of Director of Digital Technology and Artificial Intelligence at the U.S. Department of Defense. In fiscal year 2023, the department will earmark $600 million from the budget to identify key clients and systems in need of improvement.

Martell will oversee the development of U.S. armed forces’ strategies regarding AI, analytics, and data. He will also lead initiatives for joint all-domain command and control aimed at enhancing military situational awareness through technology.

Tesla on Autopilot crashes into a private plane

Tesla’s robotaxi crashed into a light aircraft as it autonomously drove to the owner’s call from the parking lot.

A Reddit user posted on the platform a clip from the surveillance camera at Felts Field in Spokane, Washington, showing the incident. In the video, a Tesla Model Y robotaxi struck a Cirrus Vision Jet worth $3.5 million, failing to recognise the aircraft’s tail section raised above the ground.

According to Electrek, the owner used the drone’s remote summon feature from the Smart Summon parking. To operate the option, the user must keep a finger on the button in the app while the robo-car is in motion. In the event of danger, simply lift the finger off the screen.

Reports: Pentagon cuts cost of HoloLens contract

The Department of Defense’s anticipated expenditure on the Microsoft HoloLens AR headset, valued at $22 billion, could be reduced by half. The military said the initial project estimate reflected the ‘worst-case cost’.

Testing of the Microsoft-integrated AR system based on HoloLens is expected to run from May 23 to June 25, 2022. Its result will be the Pentagon’s Office of the Under Secretary’s ‘Fast Field Test Report’.

Neuralink to hire smartwatch developers to build a brain–computer interface

Elon Musk invited smartwatch and smartphone specialists to join Neuralink. AI skills are not required.

In response to a Teslarati publication on Twitter, the billionaire said that over the next ten years Neuralink would tackle the problem of restoring connections between the brain and spinal cord in patients who have lost them due to injury or disease. According to Musk, the company’s brain chip could theoretically fully restore body functionality and, in the future, treat some conditions such as obesity.

The entrepreneur noted that for now the Neuralink implant grapples with ‘electronics, mechanics and software’ issues, comparable in complexity to smartwatches. Therefore the skills of specialists in this area are directly applicable, he added.

He also said the company is closely cooperating with the FDA to launch human trials of the brain chip by the end of 2022.

Google unveils method for creating multimodal models

Google unveiled the LiT method for retroactive transformation of large-image analysis algorithms into multimodal models. It is intended for developing neural networks that generate images from textual descriptions.

Neural networks using the method learn to align text with a pre-trained encoder trained on 3 billion images. According to researchers, the LiT multimodal model offers reliable image-analysis capabilities without the need for retraining for new tasks and approaches the accuracy of specialized models.

According to Google, the model trained with this method achieves 84.5% accuracy on ImageNet and 81.1% on ObjectNet without additional training. For comparison — the OpenAI CLIP model’s strongest version on ImageNet reached 76.2% and on ObjectNet 72.3%.

Startup releases AI-powered controller for garage doors

Wyze unveiled an intelligent garage door controller. The device uses an AI-powered camera and a QR code to monitor and control access to doors in the home garage.

The device connects to an existing electromechanical door opener. The controller lets users operate the door remotely via the Wyze app over Wi‑Fi.

The device uses on-camera AI to determine whether the doors are open or closed. The algorithm tracks the position of a QR code located on the door inside the garage.

The smart controller is available for $39.99, excluding shipping.

Autonomous trimaran Mayflower set sail again on a transatlantic voyage

The autonomous trimaran Mayflower has set off on a transatlantic journey from Plymouth, UK to Washington, US. The vessel is expected to cover more than 4,000 km on its own.

The vessel is equipped with four computers with Intel processors, six Nvidia Jetson AGX Xavier GPUs, two Nvidia Jetson Xavier NX boards, and several specialized chips. It also features cameras and sensors whose data are processed with machine learning software.

Computer vision algorithms and radar help Mayflower identify and avoid obstacles such as shipping containers or cargo and fishing vessels.

Earlier, the autonomous sailboat had attempted to sail from Plymouth to Massachusetts, but a fault forced it to return to the UK before the test concluded.

Intel to develop AI simulator for autonomous off‑road vehicle driving

The U.S. DARPA and Intel will develop a simulation environment for autonomous combat off‑road vehicles as part of the RACER‑Sim program. The program is planned for two years.

In the first phase, developers will create modeling platforms and map-building tools that simulate off‑road conditions. In the second phase, Intel will implement new algorithms without relying on robots, which should lower costs and shorten timelines.

Intel Labs will work with the Barcelona Computer Vision Center and the University of Texas at Austin to develop these modeling tools.

Most notable AI deals of the past week

From April 24 to 30, 2022, AI startups attracted over $250 million. Here are the most notable deals.

• Replicant, a provider of software for automating call centers, raised $78 million.
• Movable Ink raised $55 million to expand its team of software developers for marketing content personalization.
• Pyka raised $37 million to build an autonomous electric cargo aircraft.
• Baseten raised $20 million to develop technologies to ease the deployment of machine learning in business processes and products.
• CommandBar raised $19 million to develop tools for integrating smart search into applications.
• Synthesis AI raised $17 million to support its platform for creating synthetic data.
• Deepset raised $14 million.
• Loris raised $12 million to expand its team of developers building a platform for training customer-support staff to interact with customers.

Also on ForkLog:

• Elon Musk will buy Twitter for $44 billion.
• The EU will legislatively restrict the operation of algorithms in social networks.
• Meta will open a store to showcase the metaverse.
• Experts: open-source code will not solve the problems of Twitter’s algorithms.

Subscribe to ForkLog news on Telegram: ForkLog AI — all AI news!