Samsung's CRAZY AI Phone, Which Countries Are Leading in AI, AI Solves Hardest Math Problems, & More
Welcome to the AI Search newsletter. We bring you the top updates in AI news, science, & research every week.
Motionshop - Replace characters in video with 3D avatars
The framework presented in this research allows for the replacement of characters in videos with 3D avatars. The process involves a video processing pipeline to extract the background video sequence and a pose estimation and rendering pipeline to generate the avatar video sequence. By parallelizing these two pipelines and using a high-performance ray-tracing renderer called TIDE, the entire process can be completed in minutes. Additionally, the framework includes steps such as character detection, segmentation and tracking, inpainting, pose estimation, animation retargeting, light estimation, rendering using the TIDE engine, and final video composition. These steps ensure accurate replacement of characters and consistency with the original video.
Samsung unveils their AI-powered smartphone
With AI-powered transcription, summarization, real-time translation, photo/video editing, and more, Samsung’s new Galaxy S24 is jam-packed with AI features. We’ve condensed the hour-long presentation into the key points in this video. Check it out:
A simple way to reduce jailbreak attacks on ChatGPT
Researchers have developed a technique to protect language models from jailbreak attacks that can produce biased or offensive responses. They created a dataset of jailbreak prompts to test the vulnerabilities of Open AI's ChatGPT platform. The researchers then created a defense strategy called system-mode self-reminder, inspired by psychological self-reminders, which reminds ChatGPT to respond responsibly. The technique was found to significantly reduce the success rate of jailbreak attacks.
FEATURED
uPix
Generate AI selfies in just 1 click. Turn yourself into a superhero, anime character, and more!
Deepmind solves the world’s hardest math problems
Researchers have developed an AI system, called AlphaGeometry, that can solve complex geometry problems at a level approaching that of a human Olympiad gold medalist. In a benchmarking test, AlphaGeometry solved 25 out of 30 Olympiad geometry problems within the time limit, while the previous state-of-the-art system only solved 10. The system combines a neural language model with a rule-bound deduction engine to find solutions. It was trained using a large pool of synthetic data, generating 100 million unique examples. The team behind AlphaGeometry hopes that the system will advance reasoning in AI and open up possibilities across mathematics, science, and AI.
Snake-like robot 3D prints its own body to grow longer
Researchers have developed a unique robot called FiloBot that can 3D print its own body to grow longer. The snake-shaped robot has a spinning head that creates more body mass using 3D printing, allowing for elongation. FiloBot can be programmed to grow in specific ways, such as toward light or against gravity. It has a tube inside its body that feeds ink to the head, which contains a 3D printer. The head can spin at different speeds and control the amount of ink supplied, allowing for control over the direction of growth. Potential uses for FiloBot include monitoring unstructured environments, building autonomous structures, estimating the likelihood of avalanches or landslides, and testing pollution levels in hard-to-reach areas.
Robots autonomously build structures out of lattice blocks
NASA engineers have developed a system of robots that can autonomously build structures using lattice blocks. The system consists of three robots, two of which transport the lattice blocks to the construction site, while the third robot connects the blocks together. The lattice blocks are hollow and have edges similar to coat hanger wire. The robots work together to build structures, mimicking the behavior of ants in a colony. The team has successfully tested the system, building a shelter from 256 blocks that took 4.2 continuous days. This system could potentially be sent to the moon or other planets to construct structures before the arrival of astronauts.
Discovering Brand Features in Product Design Made Easy with AI
Researchers at Carnegie Mellon University have developed a deep learning architecture called BIGNet that can identify visual brand-related features in product design. The model is trained using SVG images of products and can pinpoint consistencies among curves on product images to identify the visual brand. BIGNet has been tested on different products, including cell phone brands Apple and Samsung, with a 100% accuracy rate. The researchers also assessed BIGNet on 10 automotive brands and found that luxury car makers have superior brand consistency over economy cars. The technology can save time for domain experts and has the potential for application beyond brand identity recognition.