We strive to keep our readers informed not only about events in the Bitcoin industry, but also discuss what is happening in related technology areas—cybersecurity and artificial intelligence (AI).
ForkLog AI rounds up the most important news from the AI world over the past week.
- A Microsoft Germany employee announced the imminent release of GPT-4.
- Google has revealed the date of its I/O conference.
- US senators have introduced a bill to ban TikTok in the country.
- The official stable Diffusion add-on for Blender has been released.
- Google Translate learned how to translate text on pictures.
- ChatGPT is being integrated into Slack.
- Japanese researchers have taught artificial intelligence to reconstruct images from brain activity.
- The most important AI deals of the last week.
Microsoft: GPT-4 will be released next week
On March 9, four employees from the company’s German division gave presentations at the AI in Focus – Digital Kickoff event. Andreas Braun, CTO of Microsoft Germany, casually mentioned the release of the new version of the language model.
“We’re launching GPT-4 next week. We have a multimodal model that will offer completely different possibilities, such as video,” he said.
Google I/O will be held on May 10
On May 10, Google will hold its annual I/O conference. Everyone can watch the broadcast of the event online.
Most of the announcements are expected to be specifically about artificial intelligence. A detailed schedule of meetings will appear closer to the meeting date.
In the US, a bill is registered to stop TikTok
A bipartisan group of U.S. senators has introduced a bill aimed at banning TikTok in the United States.
If passed, the Commerce Department would have the power to block foreign technology that poses a “national threat.” Services from China, Cuba, Iran, North Korea, Russia, and Venezuela may be blocked.
The administration of U.S. President Joe Biden backed the bill and called on Congress to “act swiftly.”
Meta’s large LLaMA language model has been leaked on the web. The tool’s source code appears on a 4chan image board.
The company launched the model at the end of February 2023. An approved number of users can access it.
The tech giant did not deny the leak. In addition, representatives of Meta affirmed the commitment to open models for researchers.
Stability AI releases Blender plugin
Stability AI has released the official Stable Diffusion plugin for the Blender 3D modeling application.
Stability of Blender. Data: Stability AI.
The plugin allows artists to create images directly in the program using text descriptions. You can also use existing renders, allowing you to experiment with different styles for a project without having to completely redo the scene.
Additionally, texture generation and animation are provided.
Blender Stability is completely free. It can be installed using an API key.
Google Translate learned to translate text on pictures
The web version of Google Translate has learned how to translate text in pictures.
An example of a function to translate text on an image. Source: Google Translate.
To use the feature, you have to click the “Image” button located above the text entry field and upload the file. After a while, the service returns results with translations overlaid on top of images that can be downloaded to the device. Text can also be copied.
The title under the results “Translated by Lens” indicates that the feature uses the same generative adversarial neural network used in AR Translate for Google Lens in smartphones.
ChatGPT is being integrated into Slack
ChatGPT is being integrated into the Slack enterprise messenger to help create responses to messages. This feature is available to waitlist members.
Demonstrates the AI draft feature for messages. Data: Slack.
To use a bot in Messenger, you need to click the icon with three dots in the chain, then click Create Draft.
Additionally, ChatGPT in Slack can help you find answers “on any project or topic,” as well as summarize conversations and channels.
Google tells about model development for 1,000 languages
Google engineers share their progress in creating a Universal Speech Model (USM) that supports 1,000 different languages.
According to the developers, the model has 2 billion parameters. She was trained on 12 million hours of speech and 28 billion sentences.
USM currently supports more than 100 languages. YouTube already uses this model to automatically generate subtitles.
AI taught to reconstruct images from brain activity
Japanese researchers have taught artificial intelligence to reconstruct images from human brain activity using a technique known as steady diffusion.
First, the scientists showed subjects pictures and asked them to “imagine them” during fMRI scans. The researchers then process the received signal multiple times using sophisticated algorithms.
They used the obtained results as the input of Stable Diffusion, resulting in an image close to the original image.
AI image playback results based on brain activity data. Data: Graduate School of Advanced Biological Sciences, Osaka University.
According to the scientists, however, the reconstruction of high-quality images was only possible after fine-tuning the model. This would lead to some limitations, they added.
AI Photography Studio Introducing Virtual Models
Dutch developer Danny Postma has launched a Deep Agency AI photo studio with virtual models.
The service is based on algorithms that generate images from text descriptions. Using hints, users will be able to dictate the pose and movement of the model, as well as adjust other photo settings.
A virtual model of the Deep Agency. Data: Danny Postma.
The free version provides an avatar. A paid $29 per month subscription adds three models and also opens up the ability to create Crypto twins from your own photos.
Top AI deals of the last week
From March 5 to March 11, 2023, AI startups raised over $740 million. Here are the most interesting deals.
- Anthropic, an AI startup founded by former OpenAI employees, has raised $300 million from Spark Capital.
- Marketing platform SOCi gets $120 million.
- Humane, a startup founded by ex-Apple employees, has raised $100 million.
- Overhauling AI supply chain control system raises $73 million
- Plus One, which makes order-picking robots, has landed $50 million.
- Envisics, a developer of automotive holographic systems, raised $50 million in financing.
- Fintech startup Candidly raises $20.5 million
- Artificial intelligence chip maker Mythic has received $13 million.
- Artificial intelligence fundraising platform Growfin raised $7.5 million.
- Monnai raises $6.5 million to market AI solutions for fintech companies.
what to read on weekends
What the future holds for ChatGPT, will robots replace humans, and is it ethical to use AI in warfare, read this interview with Pheon Crypto cloning startup founder Yura Fitzgerald.
Subscribe to ForkLog on social networks
Found an error in the text? Select it and press CTRL+ENTER
ForkLog Newsletter: Keeping an Eye on the Pulse of the Bitcoin Industry
Information source: compiled from FORKLOG by 0x Information.The copyright belongs to the author Богдан Каминский, and shall not be reproduced without permission