AI brain control, AI designs antibodies, realtime video games, Ernie, Pangu, Ovis-U1, Centaur, XVerse
Welcome to the AI Search newsletter. Here are the top highlights in AI this week.
Chai-2 is a revolutionary AI model that can design antibodies from scratch with unprecedented success rates, making it a game-changer for disease treatment and research. It achieves double-digit hit rates in de novo antibody design, outperforming previous methods by orders of magnitude. This breakthrough enables rapid and precise molecular engineering, paving the way for new era in medicine. Read more
Mirage is the world’s first real-time AI-powered game engine that lets players generate and modify entire game worlds live using natural language, keyboard, or controller input. It supports dynamic, interactive gameplay with photorealistic visuals and longer play sessions, demonstrated by two fully generated demos: Urban Chaos (GTA-style) and Coastal Drift (Forza Horizon-style). Read more
Baidu has open-sourced its ERNIE 4.5 model family, releasing 10 variants ranging from a lightweight 0.3 billion parameter dense model to massive Mixture-of-Experts (MoE) models. These multimodal models support text and vision tasks with a novel heterogeneous MoE architecture that shares parameters across modalities while preserving dedicated ones, improving performance on both multimodal and text benchmarks. Read more
Do you prefer to watch instead of read? Check out this video covering all the highlights in AI this week
Ovis-U1 is a powerful AI model that can understand images and text, generate new images from text, and edit images all in one framework. It's trained on a diverse mix of data and achieves state-of-the-art performance on multiple benchmarks, making it a versatile tool for various applications. Ovis-U1 is built on top of other open-source models and is released under the Apache License 2.0. Read more
Sakana AI has developed a new algorithm called AB-MCTS, which enables multiple AI models to work together and solve problems more effectively. This algorithm allows AI models to perform trial-and-error and learn from each other's strengths, leading to better performance on challenging tasks. By combining the abilities of different AI models, AB-MCTS can solve problems that were previously unsolvable by a single model. Read more
Hailuo 02 is a new SOTA video model that excels at prompt understanding, physics, camera control, and coherence. It's based on a new architecture called Noise-aware Compute Redistribution (NCR), which boosts training and inference efficiency by 2.5x, allowing 1080p video generation at a competitive cost. With unmatched efficiency and precision, Hailuo 02 is rated among the top video models in the world. Try it for free today!
Huawei Cloud has launched Pangu Models 3.0 and Ascend AI cloud services, which aim to reshape industries with AI and make large models more accessible to industry customers. Pangu Models 3.0 use a "5+N+X" three-layer architecture and are designed to focus on the practical needs of specific industry scenarios, such as finance, manufacturing, and government services. With Ascend AI cloud services, a single compute cluster can provide 2,000 petaFLOPS of compute capacity, making it easier to train and use AI models. Read more
SciArena is a new platform that evaluates the performance of foundation models in scientific literature tasks, allowing researchers to compare and vote on the best models for specific tasks. The platform has already evaluated 23 models and found that the o3 model consistently delivers top performance across all scientific domains. SciArena aims to provide a robust and trustworthy assessment of model performance and welcomes partnerships with model developers to continuously add new models to the platform. Read more
Scientists have created a brain-computer interface that allows people to control a robotic hand with their thoughts. This non-invasive technology uses electroencephalography (EEG) to decode brain signals and translate them into finger movements, enabling precise control of the robotic hand. This breakthrough could potentially help people with disabilities or injuries that affect their hand function. Read more
Centaur is an AI model that can simulate human behavior with remarkable accuracy. It was trained on over ten million decisions from psychological experiments and can predict human behavior in new situations. This AI model can help us understand human cognition and improve psychological theories. Read more
A good photo on your Linkedin or business profile makes a huge difference. You could do a physical photoshoot, which costs you over $200 and hours posing awkwardly at a camera. Or, with AI Portrait, just upload one photo, and get a portfolio of 50 professional photos in minutes. Save time and money - try it today!
A new AI-powered system can detect and quantify faults in electric motors in real-time. This system uses a combination of technologies to isolate and diagnose faults, allowing for precise severity grading and targeted protection. This can reduce maintenance costs and enhance safety in applications such as electric vehicles and critical infrastructure. Read more
XVerse is a new model by Bytedance that can generate high-quality images with precise control over multiple subjects and their attributes, such as pose, style, and lighting. It uses a novel technique called "text flow modulation" to achieve fine-grained control over individual subjects without interfering with the overall image structure. This model enables unprecedented creative control and personalization in image generation. Read more
LongAnimation is a new AI model that can generate long, colorful animations from a single reference image, with consistent colors throughout the video. It uses a special technique called "dynamic global-local memory" to keep track of colors and make sure they stay consistent. This model is a big improvement over earlier versions and could be really useful for animators and artists. Read more