Site icon

Voice assistant in space and other AI news

We strive to inform readers not only about events in the bitcoin industry, but also to talk about what is happening in related technological areas – cybersecurity and in the world of artificial intelligence (AI).

Cryplogger AI collected the most important news from the world of AI over the past week.

  • NASA will test AI technologies as part of a lunar mission.
  • Scientists have solved a secular differential equation that removes the computational limitations of “liquid” artificial intelligence.
  • The neural network was taught to decode brain waves and understand what a person sees.
  • Microsoft and Nvidia will build a supercomputer.
  • Qualcomm has unveiled the next generation AR glasses processor.
  • Robops was taught to move around any terrain.
  • The most important AI deals of the last week.

Alexa AI assistant launched to the moon as part of NASA mission

US space agency NASA successfully launched lunar mission Artemis I, in which it will test AI technologies.

An iPad-based Callisto console with an Alexa assistant and a WebEx video communication platform client installed on board the Orion unmanned spacecraft.

The mission will last 25 days 11 hours 36 minutes. During this time, the Orion spacecraft must cover about 2.1 million km.

Scientists close to discovering fast and efficient neural networks

Researchers MIT decided century-old differential equation to eliminate the computational limitations of “fluid” artificial intelligence.

In 2021, scientists have developed an AI algorithm that can learn and adapt to new information during work, and not just at the initial stage of training. These “liquid” neural networks are suitable for use in urgent tasks like monitoring a pacemaker, weather forecasting, or autonomous navigation of a robocar.

However, such models are computationally expensive as the number of neurons and synapses increases. Algorithms also require different computer programs to perform the underlying complex mathematical operations.

Now MIT has said that solving the 1907 differential equation behind the interaction of two neurons through synapses will open up a new type of fast and efficient AI algorithms.

The neural network was taught to analyze brain waves and decode vision

Research group created AI system MinD-Vis, capable of decoding brain waves and displaying what a person sees.

Comparison of the actual image and the reconstructed image based on brain waves. Data: research.

The algorithm interprets “mnemonic invariants” that form in people’s heads when they see objects. They are patterns that are stored in different memories of the same subject.

To decode the waves, the scientists used a model of self-supervised learning on the same brain activity data from several people. They also added a pre-trained AI text based diffusion image generator and a transformer neural network to these imaginary representations.

After a little fine-tuning of 1500 picture-to-picture pairs fMRIThe system was able to decode what a person sees.

MinD Vis distributed by open source. Model scales are available upon request.

Microsoft and Nvidia to build supercomputer

Microsoft and Nvidia companies concluded years of collaboration to develop a massive AI supercomputer.

The partners will host the system in the Azure cloud environment. The machine will be equipped with Quantum-2 400Gb/s InfiniBand network technology and H100 GPUs.

As part of the collaboration, Nvidia plans to use Azure Virtual Machine Instances to explore advances in generative AI and self-monitoring algorithms capable of producing text, code, images, video, or audio.

Microsoft intends to optimize the DeepSpeed ​​deep learning library for partner hardware. The company aims to reduce the amount of processing power and memory consumed during AI training workloads.

Qualcomm Unveils Next Generation AR Glasses Processor

Qualcomm announced Snapdragon AR2 Gen 1 platform for “thin and light” augmented reality glasses.

The system is a multi-chip design made using a 4-nm process technology. The Snapdragon AR2 Gen 1 supports up to nine cameras simultaneously to understand the user and the environment through a “visual analytics engine” tasked with detecting objects.

A co-processor located elsewhere on the glasses uses an AI accelerator to track gaze and perform other computer vision tasks. The third module provides connectivity to networks like WiFi 7, smartphones and PCs.

The location of the Snapdragon AR2 Gen 1 platform chips in glasses. Data: Qualcomm.

The company said the system’s multi-chip design delivers 2.5 times the AI ​​performance at half the power consumption of the XR2-based reference solution.

Previously, Qualcomm introduced the flagship Snapdragon 8 Gen 2 processor. The novelty was equipped with a dedicated AI chip, which is 4.35 times faster than its predecessor.

Waymo launched a robot taxi in downtown Phoenix

Waymo Company will allow to everyone who wants to test a robot taxi without a driver in Phoenix (USA).

Jaguar I-Pace drones will be able to deliver 24/7 deliveries to customers who have downloaded the company’s app in a designated service area.

Waymo robot taxi service area in Phoenix. Data: Waymo.

There is no information on the number of robot taxis involved.

Notion has an AI assistant for writing content

Notion Company launched testing an AI assistant for writing notes and other content.
Demonstration of the work of the AI ​​assistant. Data: Notion.

The system uses data from the Internet. The tool prompts users to choose from several templates, such as a blog post, a pros and cons list, or a to-do list. Then they need to specify the subject of the content and click “generate”. The result can be saved or redone.

The company said that for now, the feature is available to a few testers for free. Those who want to try out the tool can register on the waiting list.

In the future, Notion plans to expand the number of users and, possibly, make the tool paid.

DeviantArt introduced an image-by-text generator

DeviantArt Online Gallery launched DreamUp AI Image Generator.

The system is based on the Stable Diffusion neural network. According to DeviantArt, the tool will indicate the authors whose pictures were used for generation and tag the content created by the neural network with the AIArt tag.

Sample images generated by DreamUp. Data: Deviant Art.

The company said that users of the platform can allow or prohibit the inclusion of their work in DreamUp training datasets. Also, the authors have the opportunity to refuse to use their paintings for training third-party AI models.

DeviantArt called on other platforms to adopt this approach so that artists can share images with an audience while maintaining control over “non-human use”.

Robops was taught to move on any terrain

Researchers from Carnegie Mellon University and the University of California at Berkeley have trained a robops to move around in almost any location.

During training, experts used simulations, forcing 4,000 virtual devices to move around different terrain and gain the necessary knowledge about the world. In one day, they say, this method helped the system acquire skills that would normally take six years to master.

The researchers transferred the data collected during the simulation to the neural network and loaded it into the robot. He was able to navigate rough terrain, slippery uneven surfaces, and climb stairs.

According to scientists, the device trained in this way does not need mapping and motion planning, but adapts to conditions in real time. This can significantly reduce the cost of creating robots.

Cerebras Unveils Exascale Supercomputer

Cerebras Company developed AI supercomputer Andromeda with computing power of 1 exaflops. The machine is available for commercial and academic use.

Andromeda supercomputer. Data: Cerebras.

The system consists of 16 CS-2 servers built on a large Wafer-Scale Engine 2 chip with 13.5 million cores. For comparison: the largest supercomputer in the world Frontier has 8.7 million cores. The cluster is driven by 284 Gen AMD EPYC third generation processors.

According to the company, Andromeda cost $35 million to build, as opposed to $600 million spent on Frontier.

The most important AI deals of the last week

From November 13 to November 19, 2022, AI startups raised over $431.7 million. Here are the most interesting deals.

  • Astera Labs received $150 million to develop chips to connect AI systems in the cloud.
  • Attabotics attracted $71.7 million to expand robotic structures for vertical warehouses.
  • Creators of the technology for automating debt collection processes Symend collected $42 million
  • Spot AI attracted $40 million to develop a smart video surveillance system.
  • Speak received $27 million to expand AI language learning platform.
  • Creators of soft robotic grippers Soft Robotics collected $26 million
  • Manufacturer of smart robotic trucks Pickle Robot Company attracted $26 million
  • Owl Labs received $25 million to scale AI cameras for meetings.
  • Terzo collected $16 million to improve contract processing software.
  • Creators of embedded tools for learning the correct pronunciation of Namecoach names attracted $8 million

Also on Cryplogger:

Subscribe to Cryplogger news in Telegram: Cryplogger AI – all the news from the world of AI!

Found a mistake in the text? Select it and press CTRL+ENTER

Exit mobile version