GPT-4 teaser, a virtual photo studio and other AI news

We aim to inform readers not only about developments in the Bitcoin industry, but also about what is happening in related technological spheres — cybersecurity and the world of artificial intelligence (AI).

ForkLog AI has gathered the week’s most important AI news.

Microsoft: GPT-4 release to take place next week

On March 9, four employees of the company’s German unit выступили with a briefing at the ‘AI in Focus — Digital Kickoff’ event. Microsoft’s German CTO Andreas Braun briefly mentioned the release of a new version of the language model.

“We will present GPT-4 next week. We have multimodal models that will offer entirely different capabilities, for example, video,” he said.

Google I/O to be held on May 10

On May 10, Google will hold its annual I/O conference. The online stream will be available to all interested.

It is expected that most announcements will focus on artificial intelligence. A detailed conference schedule will appear closer to the date.

Bipartisan bill introduced in the US to ban TikTok

A bipartisan group of American senators introduced a bill aimed at banning TikTok in the United States.

If enacted, the bill would grant the Commerce Department the power to block foreign technologies deemed a ‘national threat.’ The ban could extend to services from China, Cuba, Iran, North Korea, Russia and Venezuela.

The Biden administration supported the bill and urged Congress to act swiftly.

Meta’s language model leaked online

The large language model LLaMA from Meta leaked online. The tool’s source code appeared on the imageboard 4chan.

The company presented the model in late February 2023. Access to it was granted to a vetted number of users.

The tech giant did not deny the leak. A Meta spokesperson also confirmed a commitment to openness of models for researchers.

Stability AI releases official Blender plugin for Stable Diffusion

Stability AI released the official Stable Diffusion plugin for the 3D modeling application Blender.

The add-on lets artists create images using textual descriptions directly in the program. It is also possible to use existing renders, allowing experimentation with different styles for a project without having to redo the scene.

Texture and animation generation are also available.

Stability for Blender is completely free. It can be installed using an API key.

Google Translate now translates text on images

The online version of Google Translate has learned to translate text on images.

To use the feature, press the ‘Images’ button above the text input field and upload a file. After a while, the service will return a result with the translated text overlaid on the image, which you can download to your device. There is also an option to copy the text.

The caption under the result ‘Translated by Lens’ indicates that the feature uses the same generative adversarial networks as AR Translate for Google Lens on smartphones.

ChatGPT to be integrated into Slack

ChatGPT will be integrated into the corporate messenger Slack to help draft replies to messages. The feature is available to members on the waitlist.

To use the bot in the messenger, click on the three-dots icon in the thread and select “Create draft”.

In addition, ChatGPT in Slack can help find answers on any project or topic, and summarize conversations and channels.

Google outlines progress on the universal model for 1,000 languages

Google engineers shared progress in building the Universal Speech Model (USM), supporting 1,000 languages.

According to the developers, the model has 2 billion parameters. It was trained on 12 million hours of speech and 28 billion sentences.

Today USM supports more than 100 languages. The model is already used by YouTube to automatically generate subtitles.

AI learned to recreate images from brain activity

Japanese researchers have taught an AI to recreate images from a person’s brain activity using Stable Diffusion.

First, the researchers showed participants pictures and asked them to imagine them during fMRI scanning. Then the researchers processed the signals several times using sophisticated algorithms.

The resulting data were used as input for Stable Diffusion, which generated images close to the originals.

However, according to the researchers, high-quality reconstructions could be achieved only after careful tuning of the model. This, they added, imposes certain limitations.

An AI photo studio with virtual models unveiled

Dutch developer Danny Postma opened the AI photo studio Deep Agency with virtual models.

The service is based on text-to-image generation algorithms. With prompts, users can specify the pose and actions of the model, as well as adjust other parameters of the photo.

In the free version, there is one avatar. A paid subscription at $29 per month adds three more models, and also enables creating a digital twin from your own photos.

The week’s most important AI deals

From March 5 to 11, 2023, AI startups raised over $740 million. Here are the most notable deals.

• The AI startup Anthropic, founded by former OpenAI employees, raised $300m from Spark Capital.
• The marketing platform SOCi received $120m.
• The startup Humane, founded by former Apple employees, raised $100m.
• Overhaul, the developers of an AI-based supply chain visibility, risk and compliance platform, raised $73m.
• Plus One, which builds parcel-sorting robots, raised $50m.
• Envisics, the holographic car-system developer, raised $50m.
• Fintech startup Candidly raised $20.5m.
• AI chipmaker Mythic raised $13m.
• AI-powered fundraising platform Growfin raised $7.5m.
• Monnai raised $6.5m to push AI solutions for fintechs.

Also on ForkLog:

• The number of active Bing surpassed 100 million.
• Meta reduced prices on VR headsets.
• Microsoft created a universal neural network Kosmos-1.
• AI pilot beat a human in an aerial battle in 90 seconds.
• Microsoft chief called voice assistants “dumb”.
• Fraudsters armed themselves with AI to imitate the voices of relatives of their victims.
• Users found a way to evade ChatGPT moderation.

What to read this weekend

What are the prospects for ChatGPT, whether robots will replace humans, and the ethical use of AI in war—read in an interview with the founder of the digital cloning startup Pheon, Yuri Fitzgerald.