Colab learned to write code and other news from the world of AI


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).

ForkLog AI has collected the most important news from the AI ​​world over the past week.

  • TikTok has filed a lawsuit against the block in Montana.
  • Meta taught AI to understand 4,000 living languages.
  • Tencent has launched palm recognition technology to pay for rides on the Beijing subway.
  • An image generator has been added to Photoshop.
  • AI has been taught to transform people’s thoughts into videos.
  • The writer has created 97 books using ChatGPT and Midjourney.
  • The most important AI deals of the last week.

TikTok sues Montana ban

Representatives of the TikTok service sued Montana (USA) due to the blocking of the application in the state. The respondent is Attorney General Austin Knudsen.

According to the lawsuit, the ban violates the right of citizens to freedom of speech, enshrined in the First Amendment to the US Constitution, and a number of other laws.

Google Colab will have a code generator

In the coming months, Google will add AI features to Colab for automated programming.

The cloud environment will use the Codey family of models based on PaLM 2. The algorithms have been trained on a large set of permissively licensed code snippets from external sources and tuned for the Python language.

Among the new features:

  • string generation using natural language hints;
  • code autocomplete;
  • chatbot for answering programming questions.

Initially, the update will be available to owners of a paid Colab subscription in the United States. Later, Google will make AI coding features free for the general public.

Meta introduced the Massively Multilingual Speech (MMS) family of AI language models. The project source code is available on GitHub.

MMS can identify more than 4000 spoken languages, which is 40 times more than the capabilities of existing analogues. Algorithms also taught how to synthesize speech and convert it into text in about 1100 dialects.

Available languages ​​in Meta MMS models
Available languages ​​in MMS models. Data: Meta.

To train the models, the developers used the New Testament, on the basis of which they created two datasets. The first contains audio and text in 1107 languages. The second includes unmarked speech recordings in 3809 dialects.

The company warned that sometimes MMS can incorrectly decipher certain words and phrases. There is also a lot of bias and distortion in the Bible, Meta added.

Tencent introduces palm recognition payment system

Tencent has launched a palm recognition technology to pay for trips on the Beijing subway through the WeChat Pay service.

The system was created by YouTu’s AI lab. It identifies not only the palms, but also the veins, and also takes into account the actions when the arm is extended during the scan and other characteristics. According to the developers, the technology operates in various lighting conditions.

The user needs to link the WeChat account to the recognition device and take a fingerprint in a special machine at the subway station. Then he can pay for the trip by swiping his hand over the scanner in the turnstile.

The company tested the system in Shenzhen. In the future, she plans to use the technology in offices, campuses, malls, restaurants and other places.

Bard learned to look for pictures

Chatbot Bard has learned how to display images from the Google search engine.

For example, when a user asks about the sights of New Orleans (Louisiana, USA), the chatbot will return a list of places and their photos. Each picture contains a link to the source.

Google Bard Image Search
Image search in Bard. Data: Google.

Also, the user can request images from the AI ​​directly.

Image generator added to Photoshop

Adobe has added the Generative Fill tool to Photoshop.

The technology is based on Firefly models. It allows you to scale images using text queries in English, as well as add and remove objects. The tool takes perspective, lighting, and style into account.

An example of using Generative Fill in Adobe Photoshop
An example of using Generative Fill. Data: Adobe.

The company has also added a new taskbar to the editor with a Generative Fill button that appears when objects are selected. It allows you to quickly change images even without prompts.

Generative Fill is available in the beta version of Photoshop for PC. A full release of the technology will take place in the second half of 2023.

YouTube uses AI to describe Shorts videos

YouTube has implemented the Flamingo visual language model for creating video descriptions in the Shorts section.

The algorithm developed by DeepMind analyzes the initial frames of the video and tries to understand its content. Model-generated descriptions are stored as metadata to “better categorize videos and match search results with viewer queries.”

According to Todd Sherman, director of product management for YouTube Shorts, AI text meets the platform’s responsibility standards. Situations in which the technology could make video “in a bad light” are unlikely, he added.

A DeepMind spokesperson said that Flamingo has already created descriptions for many videos, including the most viewed ones.

Scientists have created an algorithm for manipulating images

Experts from the Max Planck Institute for Informatics, MIT and Google have created the DragGAN AI algorithm for manipulating images.

The technology is similar to the Warp tool in Photoshop. Instead of simply moving pixels, it uses artificial intelligence to recreate the underlying object, according to the study.

With DragGAN, you can change the size of objects and their appearance with a couple of mouse clicks. The algorithm also allows you to rotate objects as 3D models.

According to the researchers, the approach is capable of not only creating occlusion of objects, but also deforming them according to stiffness, for example, when bending a horse’s leg.

Google expands flood forecasting tool to 80 countries

Google has expanded the availability of its Flood Hub AI flood forecasting system.

The platform can be used by governments, aid organizations and residents in 80 countries across Asia Pacific, Africa, Europe, South and Central America. Now it covers areas with a high percentage of the population at risk of natural disasters, such as the Netherlands, Vietnam, Laos, Cambodia and Myanmar.

Flood Hub uses two models to predict floods caused by overflowing rivers and streams up to seven days in advance. The system determines the amount of water entering the stream, the expected flood height and the areas that will be affected by the disaster.

AI taught to transform people’s thoughts into videos

Singaporean and Chinese researchers have developed the MinD-Video AI algorithm for video reconstruction based on brain activity data.

The system is a “two-module pipeline” of a trained fMRI encoder and a debugged version of the Stable Diffusion image generator. For her training, experts used a publicly available dataset containing videos and recordings of the brain activity of the subjects who watched them.

MinD-Video example. Data: Computer Vision and Pattern Recognition study.

According to the researchers, the algorithm reconstructs videos with 85% accuracy. The videos were found to be “reasonably high quality” as evidenced by the movement and dynamics of the scenes, they added.

The writer created 97 books with the help of AI

Fantast Tim Boucher used AI to write the AI ​​Lore book series.

The author used the ChatGPT and Anthropic Claude chatbots, as well as the Midjourney image generator. Each work contains from 2000 to 5000 words and includes from 40 to 140 illustrations.

In nine months, Boucher created 97 books. According to him, he earned almost $2,000 selling 574 copies.

The most important AI deals of the last week

From May 21 to May 27, AI startups raised over $969 million. Here are the most interesting deals.

  • Anthropic has raised $450 million to create “useful, harmless and honest AI systems.”
  • The Tools for Humanity AI team, which includes the developers of Sam Altman’s Worldcoin crypto startup, has raised $115 million.
  • A venture capital company that uses artificial intelligence to find promising startups, Moonfire received $115 million.
  • AI marketing platform Insider raises $105 million
  • Figure has raised $70 million to build a humanoid robot.
  • FlowX AI received $35 million to develop technology that allows enterprises to move legacy and new software into a single location to develop and run applications and services based on it.
  • Elementl has raised $33 million to support a Dagster-powered data orchestration platform.
  • The creators of the robotic arm and all-terrain vehicle for use in space Gitai received $ 30 million.
  • Datasembly has raised $16 million to develop a big data analytics platform for retailers and consumer goods manufacturers.

Also on ForkLog:

What to read on the weekend?

ForkLog looked at the core concepts of ethical AI and tried to figure out if this is really important for the development of safe artificial intelligence.

Subscribe to ForkLog on social networks

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

ForkLog Newsletters: Keep your finger on the pulse of the bitcoin industry!


Leave a Reply