DNA nanorobot created; TikTok unveils a DALL-E rival; and other AI news

We aim to inform readers not only about events in the Bitcoin industry but also about developments in adjacent tech spheres — cybersecurity and the world of artificial intelligence (AI).

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

Media: Sber to sell smart devices development unit

The bank Sber may sell its SberDevices smart devices development unit due to sanction-related difficulties.

According to Kommersant, the bank dropped the Sber prefix from its name, citing an expansion of its product portfolio. However, the newspaper’s sources say Sber is considering selling the division due to sanctions restricting access to foreign components and technologies.

According to the publication, amid the difficulties, staff at SberDevices are already leaving.

Nova Poshta to develop demining robots

The postal operator Nova Poshta and the Ukrainian State Emergency Service will create mine-sweeping robots.

The company plans to develop at least 40 devices to locate and neutralise mines in water and on land.

According to forecasts from Ukraine’s Ministry of Internal Affairs, pyrotechnicians will need more than 5–7 years to clear the entire country.

TSMC to launch mass production of 3nm processors

In September 2022, TSMC will begin mass production of 3nm (N3) chips. In 2023 the company will start delivering the chips to customers.

Compared with N5, the N3 technology will offer a 10–15% higher performance at the same power and complexity, and reduce chip power consumption by 25–30% while maintaining frequency. It will also increase transistor density by about 1.6x.

The FinFlex technology used in N3 will allow designers to mix and match different types of standard cells within a single block for precise optimization of performance, power consumption and area. It will be useful for complex products, including CPUs and GPUs.

In the future TSMC will launch process nodes N3E, N3P and N3S. The former will optimize N3, and the others will improve N3E for various application classes.

Tesla owners urged not to test Autopilot on children

A Tesla enthusiast suggested testing the Full Self-Driving (FSD) system on children. The Verge journalist urged against doing so.

The U.S. National Highway Traffic Safety Administration (NHTSA) also warned Tesla owners against testing the system on children.

Regulators’ attention was drawn to videos showing users testing FSD on minors: placing them on a road or asking them to cross in front of the vehicle.

According to the video authors, the drone correctly recognized children of different heights and slowed down.

The NHTSA urged not to risk lives to test automotive technologies. Consumers should not recreate testing scenarios in real life or involve others in them, the agency added.

Earlier, The Dawn Project tested the technology on child-sized dummies. A robotic car moving along a straight road at 40 km/h failed to recognise them and struck.

Google updates Search to curb clickbait

In the coming months, Google will update Search to improve result quality.

According to the developers, the changes will reduce the ranking of clickbait headlines from aggregator sites to attract traffic via search engine optimization. Instead, the updates will elevate original and authentic reviews created by “people for people”.

The company will start rolling out the “Helpful Content Update” to English-speaking users on August 22. Changes to simplify finding high-quality and original product reviews are expected in the coming weeks.

TikTok unveils text-to-video background generator

TikTok has introduced an AI greenscreen background generator for video from text. The tool is available to Android and iOS users.

Compared with the photorealistic outputs of OpenAI’s DALL-E 2, the AI greenscreen algorithm generates primitive abstract images. According to The Verge, TikTok may have done this deliberately to avoid risks of violating platform rules such as nudity or violence.

Google teaches office robots to converse

Google has taught Alphabet’s office robots to understand natural-language commands.

The developers equipped Everyday Robots prototypes with PaLM. Now, if a robot hears the phrase “I spilled my drink, will you help?”, it will analyse it through an internal list of potential actions and interpret it as “bring me a sponge from the kitchen”.

Google says that the PaLM-SayCan system allowed devices to plan correct responses to 101 prompts in 84% of cases, and to execute them in 74%.

Stability AI unveils free text-to-image generator

The startup Stability AI has provided access to more than 1,000 researchers to its text-to-image generator Stable Diffusion. A public launch of the tool is planned in the coming weeks.

The system runs on high-performance hardware and generates 512×512 pixel images in seconds for any text prompts.

The company says the algorithm can create images featuring public figures, nude people and even toxic content, which OpenAI’s DALL-E 2 does not do due to its filters.

Researchers create a DNA nanorobot

Researchers at the University of Montpellier have built a DNA nanorobot. The device, the size of a human cell, will help study viruses and pathogens at a microscopic level.

Researchers used the DNA origami method to assemble the nanorobot from three DNA structures. They then connected it to a mechanoreceptor-recognition molecule, and learned to activate it.

This helped scientists determine when key signaling pathways are activated at the cellular level under applied force. In the future, the same could be done with other biological and pathological processes.

The researchers said this is the first man-made, self-assembling DNA-based object capable of applying and controlling force with a resolution of one piconewton.

However, the researchers noted the robot is vulnerable to degrading enzymes, so they need to figure out how to modify its surface and address this vulnerability.

The United States’ first fully robotic restaurant to open

On August 28, 2022, Mezli will open in San Francisco, USA, a healthy-eating restaurant operated by robots with no human involvement.

The devices can portion ingredients, place them in paper bowls, heat them, and hand them to customers. The company says the restaurant can process about 75 orders per hour.

Besides ready-made dishes, customers can assemble a plate from their chosen ingredients. There are about 64,800 possible combinations.

Most notable AI deals of the past week

From August 14 to 20, 2022, AI startups raised more than $378 million. Here are the most notable deals.

• SAIC Mobility raised $148 million to scale its robotaxi service in China.
• Pliops raised $100 million.
• GrowthSpace raised $25 million to expand the platform globally.
• HyperTrack raised $25 million to advance software for last-mile logistics management.
• BigPanda raised $20 million.
• Jiminny raised $16.5 million to build a conversational analytics platform.
• Sync Computing raised $15.5 million to support automatic cloud resource optimization.
• Mobot raised $12.5 million to expand its fleet of robots for testing mobile apps.
• CleanRobotics raised $4.5 million.
• Regression Games raised $4.2 million to develop the platform and ecosystem for using AI in gaming and esports.

Also on ForkLog:

• Hyundai will create an AI and robotics institute.
• Chinese tech giants have handed authorities data on algorithms.
• Oracle will conduct an algorithm audit of TikTok.
• British hospitals lost $18 million due to an AI startup.
• Media: the algorithm randomly laid off 60 Facebook moderators.

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