Matthias Bastian

Dec 13, 2023

Midjourney breaks out of Discord: Image generation on the web is available in an alpha version. To use the alpha version, you must have generated at least 10,000 images and log in to the current alpha website. Compared to Discord, it offers a simpler interface where you can set parameters like aspect ratio, chaos, or style using sliders and tags. You can also insert images into a prompt with a single click. X user Nick St. Pierre shows the new Midjourney site in action in a video below. There is no news yet about the upcoming version v6, which Midjourney announced in September for this year. The new version is supposed to follow the prompts more closely, which is currently the biggest advantage of OpenAI's image generator DALL-E 3.

Video: Nick St. Pierre

Comment

Matthias Bastian

Dec 13, 2023

AI in practice

The New York Times has hired Zach Seward, co-founder of Quartz, as its first editorial director for AI initiatives. Seward will work with newsroom leadership to establish guidelines for the use of generative AI in news production. He is also expected to build a small team to experiment with AI tools, prototype ideas, and design AI training programs for journalists. The move comes as news organizations are cautiously exploring AI tools for tasks such as automating publishing and generating headlines. AP has an official partnership with OpenAI. The Times has already begun allowing newsroom staff to experiment with AI tools. But concerns remain about the potential impact on quality, jobs, and transparency, as recent examples show.

Comment Source: Wall Street Journal

Matthias Bastian

Dec 12, 2023

Update

AI in practice

Run LLMs on your M Series with Apple's new MLX machine learning framework

Matthias Bastian

Dec 12, 2023

AI research

Microsoft puts GPT-4 ahead of Gemini Ultra again, using Google's own tricks

A vividly colored, hand-drawn illustration in a 16:9 aspect ratio, showcasing two large language model chatbots racing against each other. The scene is lively and full of color, emphasizing the theme of language and AI. One chatbot is depicted as a whimsical robot with elements like alphabets and script flowing around it, symbolizing its language processing capabilities. The other chatbot is more abstract, surrounded by colorful digital patterns that resemble word clouds or streams of text, representing its data processing power. The background is a vibrant mix of colors, with hand-drawn elements that include floating words, sentences, and linguistic symbols, creating a dynamic and playful atmosphere. This composition clearly conveys the idea of language models racing in a world of words and digital language.

Matthias Bastian

Dec 12, 2023

AI in practice

Two more publishers caught publishing AI articles without proper disclosure

The Arena Group, publisher of Sports Illustrated, has fired its CEO Ross Levinsohn. This follows a scandal involving AI-generated articles and fake writers.

Matthias Bastian

Dec 11, 2023

AI in practice

YouTuber "Greg Technology" has recreated Google's discredited multimodal Gemini AI demo using OpenAI's GPT-4 Vision to demonstrate real-time voice and vision prompts. The original Gemini AI demo video, which was criticized for being staged and not recorded in real-time, featured voice interactions that were later dubbed in. In response, Greg Technology released a video using GPT-4V in which he discussed a drawing, asked about emoticons, and had the AI identify a game. It's not as polished as Google's demo, of course, but it's real-time and real. Greg has published his demo code on GitHub.

Comment Source: YouTube