This Wired article tells the story behind the development of the “Transformer,“ a revolutionary AI architecture that powers modern language models like ChatGPT. A team of eight Google researchers developed the Transformer in 2017, based on the concept of attention. The team worked hard to complete the paper before an important conference deadline, and their …
The British startup has popularized the “diffusion“ technique for AI image generation developed by Munich students. Now Robin Rombach, Andreas Blattmann and Dominik Lorenz have left the company, three of the original five people involved in the research project. Source: Forbes
It seems the startup doesn’t have much time to control the individual chatbots in its store, as TechCrunch shows.
Researchers from Washington and Chicago have developed “ArtPrompt“, a new method to bypass security measures in language models. Using this method, chatbots such as GPT-3.5, GPT-4, Gemini, Claude, and Llama2 can be tricked into responding to requests they are supposed to reject using ASCII art prompts. This includes advice on how to make bombs and …
Amazon Web Services (AWS), Accenture, and AI startup Anthropic (makers of Claude) are joining forces to help organizations in highly regulated industries, such as healthcare, government, and banking, deploy customized AI models quickly and responsibly. The partnership will enable organizations to access Anthropic’s AI models, including the entire Claude 3 family, through AWS’ Bedrock platform. …
Amazon shows a small, nice example of a practical AI application: Their assistant will generate an Amazon product page from a link to a product in a merchant’s own online store, reports The Verge. The service is initially available in the US.
Stable Video 3D creates 3D models from a single photo. It can be used for free for non-commercial applications.
Google researchers show VLOGGER, which can create lifelike videos of people speaking, gesturing and moving from a single photo. This opens up a range of potential applications, but also raises concerns about forgery and misinformation. Source: VentureBeat
Pika already offers a “lip sync“ feature that makes people speak in generated videos. Now there is an option to add sound to a generated clip, such as background noises and effects. Source: VentureBeat
While many AI videos are only a few seconds long, Story.com allows up to 1 minute. A storyboarding feature is supposed to ensure that the clips ultimately meet the user’s ideas and needs.