Tesla cuts Autopilot developers; AI learns to play Minecraft, and other AI news

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

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

IBM, Cisco and Nvidia halt license sales in Russia

IBM and Cisco have ceased renewing licenses for using equipment in Russia. This will block activation of new devices, as well as support and maintenance for those already in use.

According to «Ведомостей», the issue will affect data-storage systems based on processors with the IBM Power architecture and Cisco networking equipment.

Nvidia has also halted sales and renewal of software licenses in Russia. The restrictions affected developers of popular cloud gaming services that rely on the company’s software to optimise GPU performance.

In the future, the issue could affect owners of supercomputers such as Yandex and Sberbank.

Tesla cuts Autopilot developers

Tesla laid off 200 employees from the Autopilot development team. The company also closed its office in San Mateo, California.

According to Bloomberg, many of the laid-off workers worked on evaluating and labeling Autopilot data received from customers. The remaining part of the 350-strong team was moved to another office.

Cruise’s autonomous taxis paralyze traffic in San Francisco

More than six Cruise robotaxis blocked traffic in San Francisco for several hours.

According to TechCrunch, on June 28, 2022 several autonomous vehicles suddenly stopped working in the Fillmore neighborhood. Cruise staff arriving to the scene had to manually move the robo-cars.

The company apologised for the incident but did not explain the reasons for such behaviour of the autonomous vehicles.

Neural interface helped a paralyzed man eat with robotic arms

Researchers from Johns Hopkins University unveiled a neural interface that helps a paralyzed man eat with robotic arms.

The method, built on a shared control system, minimizes mental effort required to perform tasks. The technology enabled a partially paralyzed person to map four movements to 12 degrees of freedom of the manipulators.

The participant made light hand gestures in response to system prompts such as “select the cut location,” so robotic arms with a fork and knife could divide the food and bring it to the mouth. According to researchers, the man was able to eat a dessert in 90 seconds.

In the future, researchers plan to add tactile feedback akin to touch, as well as improve the system’s accuracy and efficiency.

Startup tested AR lenses on a person

Mojo Vision tested the Mojo Lens AR contact lenses on a human.

The lens features a green monochrome MicroLED display with a diameter of less than 0.5 mm and a resolution of 14,000 pixels per inch. The device is equipped with an ARM Core M0 processor and a 5 GHz radio module with ultra-low latency. It also includes eye-tracking and image-stabilisation sensors.

Mojo Lens houses a micro-battery integrated into an external ring, which in the final version could last a full day. Control of the gadget is via eye movements.

The company announced plans to begin testing the lens on other volunteers, whose feedback will feed into the commercial version of the product for licensing and sale.

Autonomous Mayflower arrives in Plymouth, Massachusetts

The autonomous vessel Mayflower reached the shores of Massachusetts (USA).

The trimaran was towed into Plymouth Harbor and moored alongside the original Mayflower, the ship that carried pilgrims across the Atlantic in 1620.

Developers say they will now study the data collected by the drone and the performance of the AI systems.

Neural network trained to play Minecraft

OpenAI engineers trained AI to play Minecraft. The neural network can run, swim, navigate obstacles, gather resources, craft diamond tools, hunt animals, and cook food.

The researchers developed a method of pretraining on videos that enables using a large corpus of unlabeled data. They also trained the AI to play using mouse and keyboard emulators.

Initially, engineers showed the neural network 2,000 hours of labeled Minecraft gameplay with user-keypress indicators. They built a predictor-clicks algorithm that labeled another 70,000 hours of gameplay from open sources.

Subsequently, researchers asked participating gamers to create a new world in Minecraft, collect starting resources, and craft essential items. The AI learned to build a crafting table, fabricate items, and construct shelters.

They also used the RL-method to train the network to craft diamond tools.

Niantic and NBA to create AR basketball game

Niantic and the National Basketball Association (NBA) will create the mobile AR game NBA All-World. The app will be available for iOS and Android.

According to Polygon, players will have to “explore their surroundings” in search of some league stars, including Chris Paul, Stephen Curry and James Harden. Players will be able to outfit characters and upgrade teams with items found at real-world locations.

Players will also be able to challenge virtual teams in one-on-one matches at various real-world locations, including basketball courts.

Enthusiasts combine Wordle and DALL-E mini into Wordalle

Members of the Huggingface Projects community created a web game Wordalle. Players must guess the prompt from a generated AI image.

Developers merged the Wordle puzzle with the Craiyon neural network (formerly DALL-E mini). Players must place letters in cells: if correct and in the right place, the tile turns green; if the player misplaced the letter, it turns yellow.

Users have seven attempts to guess the algorithm-generated prompt. Words can also be skipped.

According to developers, prompts used in Wordalle are varied but easily guessable, such as “Yoda doing yoga” or “pug-taco.”

The startup developed a facial-recognition pet door

The company Petvation developed a facial-recognition pet door.

The system is equipped with infrared cameras with a 120-degree field of view and IR illumination, a downward-facing video sensor, and motion sensors. The machine-learning algorithm identifies cats and dogs, as well as other animals such as raccoons, squirrels, and skunks.

According to Petvation, the system does not require the pet to look into the camera before exiting. The engineers also mitigated the risk of pinching a paw or tail in the door, the company said.

Most notable AI deals of the past week

From June 26 to July 2, 2022, AI startups raised over $230 million. Here are the most notable deals.

• Gloat raised $90 million to support an AI job marketplace.
• Speechmatics raised $62 million to expand inclusive speech-to-text technology.
• Modular raised $30 million.
• Mona raised $14.6 million to develop a metaverse platform for 3D engineers.
• Industrial Next raised $12 million to sharpen manufacturing technologies for EV suppliers.
• Kogniz raised $10 million.
• AI Squared raised $6 million to help companies integrate AI into existing applications.
• Heyday raised $6 million to build a knowledge base from services you already use.

Also on ForkLog:

• Cerebras broke a record in training an AI model on a single device.
• An engineer created an AI tool for identifying Holocaust victims.
• FBI: scammers use deepfakes in online interviews.
• In the US, federal rules for drones are being developed.
• Scientists created an algorithm for predicting crimes a week in advance.

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