Algorithm Develops a Processor, Robotic Dog Learns the Command ‘Aport!’ 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 technology spheres — cybersecurity, and now the world of artificial intelligence (AI).

\n\n\n\n

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

\n\n

\n\n

Medvedev Signals Possible Ban on Certain AI Forms

\n\n

Some forms of artificial intelligence could be banned in Russia on ethical grounds. Deputy Security Council Chairman Dmitry Medvedev said.

\n\n

“We need to form a position on sensitive issues, I mean […] a possible ban on the use of certain forms of artificial intelligence, if this becomes necessary, restrictions on its use in certain areas,” he said at a meeting on protecting Russia’s national interests in AI development.

\n\n

According to Medvedev, the main thing is not to overstep the mark. He noted that carelessness could lead to banning technologies that are important for human development.

\n\n

China Deploys AI to Track UFOs

\n\n

China uses artificial intelligence to monitor UFOs.

\n\n

According to Chen Li, a Air Force spokesman for the PRC, “the AI’s ‘nonstandard thinking’ will help explain the nature of unidentified flying objects in the sky and unite disparate incidents surrounding their appearance. He believes AI can determine what the aerial object really is — an enemy aircraft, a natural phenomenon, or something unknown to science.”

\n\n

Despite a multitude of conspiracy theories around UFOs, Chinese scientists and authorities say the objects have artificial and earthly origins.

\n\n

Apple Announces a Series of AI Features at WWDC 2021

\n\n

Apple held the Worldwide Developers Conference 2021, during which it announced several new AI-based features and tools for developers.

\n\n

Siri was taught to process user voice requests on-device without internet access. Developers said the processing is handled by neural networks. They argued that offline recognition improves privacy and enhances the user experience.

\n\n

Siri will also get new voices and appear on devices from third-party manufacturers.

\n\n

Apple also announced an update to RealityKit for building augmented reality apps. With it, developers can generate 3D models from photos taken with iPhone.

\n\n

\n\n

To do this, they need to take a series of pictures of the object from all sides, including from below, using any camera. Using the Object Capture API in macOS Monterey, you specify the folder with the source images, select the level of detail, and run the rendering process.

\n\n

According to the company, the ready-made models are optimised for the AR Quick Look platform, which lets you add 3D objects to apps or websites on iPhone and iPad.

\n\n

Facebook Announces a Contest to Teach AI Agents to Master NetHack

\n\n

Facebook announced a contest to create an algorithm capable of mastering NetHack — one of the most challenging video games in history.

\n\n

Organizers do not expect participants to achieve high scores. The task is to develop AI agents that can earn as many points as possible and learn to navigate complex environments with minimal compute.

\n\n

The competition runs until October 15, and its results will be announced at the NeurIPS 2021 conference in Australia.

\n\n

NetHack is a classic 1987 roguelike RPG. It presents a difficult, dynamic, and unpredictable world of mazes, each generated randomly.

\n\n

Yandex Develops YaML Language Model with 3 Billion Parameters

\n\n

Yandex developed the YaML language model and created the Zeliboba service to demonstrate its capabilities.

\n\n

\n\n

The company said the model contains up to 13 billion parameters. It was trained on terabytes of information, including pages from search and Wikipedia, news notes and social media posts.

\n\n

“Zeliboba” uses a 3-billion-parameter variant. Its task is to generate each subsequent word in a sentence started by the user.

\n\n

The service was available to everyone for several hours, but has since been closed. Yandex admitted they launched the technology too early by mistake.

\n\n

Enthusiasts Build a Free GPT-3 Alternative with 6 Billion Parameters

\n\n

A group of EleutherAI enthusiasts released the open-source language model GPT-J. The developers say it runs at roughly GPT-3 level for a model of similar size.

\n\n

GPT-J contains 6 billion parameters. It was trained over five weeks on 400 billion tokens from The Pile — an 835 GB collection from 22 smaller datasets, including Arxiv, PubMed, Wikipedia and others.

\n\n

The GPT-J source code is available on GitHub.

\n\n

Tesla Autopilot Registered Phantom Traffic Signals at 130 km/h

\n\n

Social media users were concerned by a video showing Tesla allegedly in autopilot mode recognizing phantom traffic signals while traveling on a highway at 130 km/h.

\n\n\n\n

The system responded to a van in front of the car carrying devices in its cargo area. The video author stressed that Tesla was not in autopilot mode.

\n\n

The author posted this video in the comments. It’s a truck full of traffic lights. pic.twitter.com/h2LwyL65ck

— FSD in 6 months (@FSD_in_6m) June 2, 2021

\n\n

The system has faced criticism for false positives when recognizing traffic signals. According to user feedback, the company has not yet resolved this issue.

\n\n

@elonmusk WTH? pic.twitter.com/MMFGG22X87

— blunt_cognition 🪑 (@SH_LoSHka) October 23, 2020

\n\n

Boston Dynamics Teaches Robotic Dog to Respond to the ‘Aport!’ Command

\n\n

Boston Dynamics released instructions for training the Spot robot dog to respond to the command \”Aport!\”

\n\n

According to the tutorial, the user must take at least 400 photos of a rope toy in a given environment from different angles. For each new environment, the image collection and training process must be repeated.

\n\n

\n\n

As a result, the robotic dog slowly approaches the toy, grabs it with a gripper, turns, and tosses it to the owner.

\n\n

China Unveils Budget Robotic Dog That Follows Its Owner and Carries Water

\n\n

China’s Unitree Robotics unveiled the Unitree Go1 robotic dog.

\n\n\n\n

According to the developers, the device is not meant for commercial use. The dog-cost companion is better suited as a follower that carries water on runs or groceries in a supermarket.

\n\n

Go1 is equipped with an advanced navigation system and can overcome obstacles. It carries 10 cameras and sensors, managed by a 16-core processor with a video accelerator delivering 1.5 teraflops of performance.

\n\n

The companion costs $2,700, making it one of the most affordable on the market. By comparison, Boston Dynamics’ Spot starts at $74,500.

\n\n

Developer Creates AI Service to Assess a User’s “Normality”

\n\n

A Dutch developer, with EU support, built a service that determines how “normal” a person is from a photo.

\n\n

On the site, several algorithms attempt to assess attractiveness, age, gender, body mass index, and remaining years of life from facial features.

\n\n

In this way, the developers want to illustrate how companies use facial recognition for extended analytics and how algorithms judge people by appearance.

\n\n

Also on ForkLog:

\n\n

• IBM will establish in the United Kingdom a centre for quantum computing and artificial intelligence.
• Experts: European AI rules will open the doors to mass surveillance.
• DeepMind: for creating AGI reinforcement learning is enough.
• Google accelerated chip development from months to six hours using AI.
• OpenAI softened bias and toxicity in GPT-3.
• Apple introduced passwordless authentication using Face ID and Touch ID.

\n\n

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