Nuclear-threat detection, Meta layoffs, and other AI news

We aim to inform readers not only about developments 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 gathered the most important AI news from the past week.

Meta AI team hit by mass layoffs

Meta fired a team of researchers in the machine-learning infrastructure stack.

According to a former employee, the Probability group had 50 people, not counting managers. 19 of them were engaged in Bayesian modeling, 9 in ranking and recommendations, 5 in increasing technology efficiency, and 17 in AI for chip design and compilers.

Earlier Meta announced layoffs of 11,000 workers. After that, Meta CEO Mark Zuckerberg hinted that mass layoffs would not affect AI/ML researchers.

Insider: Apple will drop the word “Hey” from the trigger phrase Siri

Apple plans to drop the word “Hey” from the trigger phrase for Siri.

According to Bloomberg journalist Mark Gurman, engineers have been working on this change for several months. Teaching Siri to understand a single wake word is technically challenging, requiring substantial AI training and basic engineering work, he added.

Currently developers are testing the update on employees. One task is to train Siri to understand activation words across multiple dialects and accents.

It is expected to roll out in 2023 or 2024.

Amazon unveiled a new warehouse robot

Amazon at Delivering the Future conference unveiled the Sparrow robotic manipulator for repetitive warehouse tasks.

The device uses computer vision and AI to move items before packing and shipping to customers. The robot can lift and sort objects of various shapes and sizes.

According to the company, the system can identify about 65% of Amazon’s inventory.

GitHub shows off voice assistant for coding

The GitHub service launched testing of a voice assistant for the AI coding system Copilot.

The Next team of engineers and researchers is developing the technology. According to them, the assistant can already write and edit a code fragment, switch to another method, run a program, and explain what a particular line does.

The project is experimental, and it is not yet clear whether the voice assistant will become a full-fledged product.

Waymo can charge for fully driverless delivery in California

Authorities in California allowed Waymo to charge for operating driverless robot taxis in San Francisco.

The Department of Motor Vehicles of the state updated a permit for commercial use of autonomous and driverless operations. Now Waymo can charge for operating robocars without a safety driver for services such as food and grocery delivery.

For a fully driverless taxi deployment, the company must obtain a permit from the California Public Utilities Commission.

AI to help Ukrainians find jobs

In Ukraine they will deploy AI to help citizens find employment and start businesses.

According to representatives of the State Employment Service, the technology “quickly and accurately” will provide job seekers with a pool of professions based on their indicated abilities and flexible skills, which employers pay the most attention to during hiring.

Also the system will automatically generate an interactive resume. The agency stated that it will give job seekers an edge because it will contain all the information recruiters need.

When a tendency towards entrepreneurship is identified, AI will offer a comprehensive training in entrepreneurship. Upon completion, the system will generate a named certificate with a unique number.

Research: AI could speed up nuclear-threat detection

Experts at the Pacific Northwest National Laboratory (PNNL) say that AI can accelerate detection of nuclear threats.

According to the PNNL report, data analysis and machine learning methods can help experts IAEA uncover potentially malicious activity. To this end, researchers are developing several AI projects. Among them:

• a machine-learning model to detect leakage of nuclear materials at processing facilities;
• a system based on the Google BERT language model capable of analyzing text and identifying signs of nuclear weapons proliferation;
• a tool that can determine the origin of nuclear materials from images.

According to the PNNL, these algorithms can already optimize the process of detecting nuclear threats. However, it will take some time before agencies like the IAEA deploy AI in their work, the researchers added.

Boston Dynamics marks 30 years

The company published a video showing the evolution of robotics over the past 30 years.

The video begins with a demonstration of next-generation dynamic one-legged, bipedal and four-legged devices created by the company’s founders at Carnegie Mellon University (1980–1986) and MIT (1987–1995). The video also shows the evolution of Spot, Atlas and other robots.

According to Boston Dynamics, the video is dedicated to 30 years of innovations, research, and collaboration.

Midjourney releases a new version

Midjourney image-generation service updated to version four.

According to project founder David Holz, the algorithm received a new codebase, architecture and “knowledge,” including about locations and beings. The neural network can correctly recognise fine details, understand more complex prompts and handle scenes with multiple objects and characters.

Developers also improved the function of blending two images. Now Midjourney more effectively composes images, merging colors, composition, motifs and styles in a 50/50 ratio.

Holtz noted that since Midjourney V4 can generate more realistic illustrations, moderators will start enforcing community rules more strictly.

Researchers develop algorithm to improve night photos

Researchers at the University of California, Los Angeles developed AI VEViD algorithm for enhancement of photos.

The system simulates physical processes and can adjust low-contrast or low-light images. VEViD enables neural networks trained on high-quality images to recognise objects in night-time photos without retraining.

According to researchers, the algorithm is high-performance, interpretable and does not require labeled data for training. It can be used for processing images in autonomous driving tasks, night-time surveillance and space science studies.

The week’s key AI deals

From November 5 to 12, 2022, AI startups attracted over $98.5M. Here are the most notable deals.

• The Eliyan chiplet interconnects tech creators attracted $40M.
• Mem received $23.5M to expand an AI-powered workspace for automatic note organization.
• Veriti raised $18.5M to improve a platform for unifying cybersecurity infrastructure.
• Movio developers got $9M for create video with talking avatars.
• Anything World raised $7.5M to advance AI tech for 3D model animation.

Also on ForkLog:

• Elon Musk fired the Twitter ethics AI team.
• Nvidia slowed the A100 chip to circumvent anti-China sanctions.
• Google taught a robot to program itself.
• IBM unveiled 433-qubit quantum processor.
• Intel unveiled Max-series AI chips for supercomputers.
• KFC bot congratulated customers in Germany on the Kristallnacht anniversary.

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