We aim to inform readers not only about events in the bitcoin industry but also about what happens in adjacent technology spheres — cybersecurity and the world of artificial intelligence (AI).
ForkLog AI has gathered the most important AI news from the past week.
- Since July 2021, 392 robomobiles have been involved in crashes.
- Tesla and Samsung signed a multibillion-dollar deal to supply cameras for autonomous driving.
- Adobe developed a tool to restore damaged photos in Photoshop.
- Facebook collects data on people seeking abortion-related services.
- An artificial algorithm is used to analyse images from the James Webb Space Telescope.
- Researchers trained an algorithm to create videos from textual descriptions.
- The most important AI deals.
Around 400 robomobiles have been involved in crashes in the past year
The National Highway Traffic Safety Administration (NHTSA) published a report on crashes involving robo-cars and driver-assistance systems.
According to the agency, from July 1, 2021 to May 15, 2022, 392 crashes involved autonomous vehicles. In 108 cases another vehicle was involved, and in 11 cases cyclists and pedestrians were involved. Of the 98 crashes with injuries, 11 involved serious injuries.
273 crashes involved Tesla robo-mobiles using Autopilot or Full Self-Driving. Five incidents were fatal.
Honda vehicles were involved in 90 crashes, Waymo 62, Transdev Alternative Services 34, Cruise 23, Subaru 10. Autonomous vehicles from other manufacturers such as Ford, General Motors, Volkswagen and Toyota were involved in five or fewer crashes.
Tesla to buy cameras from Samsung for robo-mobiles
Tesla has formed a partnership with Samsung to supply cameras for robo-mobiles.
According to Electrek, the tech giant won a tender valued at $3.2-$4 billion. Samsung will supply Version 4.0 cameras with a 5 MP resolution, delivering five times sharper image than the 3.0 modules.
Tesla will integrate these video devices into all robo-mobiles over the next few years, according to the source.
Adobe unveils tool for restoring old photographs
Adobe has developed a neural filter to restore damaged old photos in Photoshop.
The system runs on Sensei AI. The neural filter cleans images of grime, scratches, creases and tears. The technology smooths human skin in portraits, removing unwanted textures or noise and adding tones.
The filter can work alongside other tools such as Content-Aware Fill to restore missing parts of a photo.
The company also announced a free web version of the photo editor with basic features.
Meta develops algorithm to model 3D avatars from selfies
Meta will unveil a method to create realistic 3D avatars using a smartphone at Siggraph 2022.
The approach allows digitising a head in four minutes, then generating its 3D model with a wide range of expressions. To do this, users would need to ‘scan’ the face with the front camera of a smartphone, recreating 65 specific expressions.
Researchers say the process of generating a 3D avatar on a high-performance computer would take about six hours. Rendering can also be done in the cloud.
The method does not work with glasses and struggles to render long hair, the researchers add.
Study: Facebook tracks people seeking abortion services
Facebook’s advertising tools track people seeking abortion services. The system should filter confidential health data, but instead it collects it, Reveal said.
Researchers found hundreds of crisis pregnancy centers use the Meta Pixel analytics tool. The system allows organisations to track site visits to tailor ads on Facebook.
According to the report, crisis pregnancy centers are often not licensed medical facilities and are not affiliated with HIPAA. They are mainly run by religious organisations that urge people to abstain from abortion.
According to Reveal, Meta Pixel collects “extremely” confidential information, including scheduled appointments. External marketing firms opposing abortion could also access the data collected by the tool, even if their sites were not visited.
ByteDance to invest in VR content development
The TikTok-owner ByteDance plans to invest “a lot of money” in VR content development.
According to Protocol, most of the funding will go toward hiring staff for the company’s VR headset maker Pico. Most roles are in the Pico Studios division, focused on content.
Microsoft Teams uses AI to improve audio in video conferences
Microsoft has deployed AI to improve audio quality in Teams video calls.
The company says the new algorithms can minimise echo and reduce interruptions by removing overlapping reflections. All processing is performed locally on client devices, Microsoft added.
According to The Verge, the developers built the technology using 100,000 modelled rooms and trained it on 30,000 hours of speech. The tech giant also paid Teams users to record their voices to train AI to recognise thousands of different devices.
The feature ships with other updates, including AI-assisted bandwidth management during video calls and improved text rendering.
AI will study James Webb Space Telescope images
AI will be deployed to study the first images from the James Webb Space Telescope (JWST). The images will be released on July 12, 2022.
A team of 50 researchers will apply the Morpheus AI system, trained to analyse pixels, find blurred objects in space images and determine whether the structures are galaxies and what type. The model will also help study the growth of dark matter and the birth of stars.
Morpheus was trained on 7,600 images of galaxies captured by NASA’s Hubble Space Telescope. Researchers however plan to retrain the algorithm to work with JWST data.
Researchers teach neural network to create video from text description
Researchers at Tsinghua University (China) developed a 9‑billion-parameter CogVideo transformer for converting text to video. The network currently handles only Chinese.
The researchers trained the algorithm by inheriting the pre-trained image generator from CogView2.
The model creates a 4-second video at 480×480 pixels with 8 frames per second.
New Voicemod filter lets you speak with Morgan Freeman’s voice
Voicemod has deployed AI to use Morgan Freeman’s voice in real time. The beta of Voicemod AI Voices also includes seven new filters.
Developers enlisted native-English-speaking professional voice actors with the necessary speech characteristics to build a training dataset. Sound engineers then processed the data with traditional methods such as adding background music and dynamic effects.
Voicemod AI Voices is available on PC. The filters can be used on Twitch streams or Discord calls.
Key AI deals of the past week
From June 12 to 18, 2022, AI startups attracted over $237 million. Here are the most notable deals.
- Invoca raised $83m to develop software for analysing call-centre operator performance.
- Charm Therapeutics raised $50m to advance its AI platform for modelling complex protein interactions.
- Lightning AI raised $40m to expand the team of developers building AI tooling.
- Alice Technologies raised $30m.
- Able raised $20m to accelerate data processing for banks and other lenders.
- Ahana raised $7.2m.
- Workstream raised $7m to develop its platform for optimising enterprise analytics workflows.
Also on ForkLog:
- Google sidelined an engineer, who claimed AI possessed consciousness.
- The Chinese government used anti-COVID infrastructure to quell protests.
- Experts doubted the emergence of consciousness in Google’s algorithm.
- Tech giants joined the fight against misinformation in the EU.
Subscribe to ForkLog AI News on Telegram: ForkLog AI — all the news from the AI world!