Advances in AI Across Video Creation, Image Generation, Chatbots, and Multimodal Benchmarks

Welcome to our weekly newsletter ??, your go-to source for the latest developments and trends in Generative AI.

Each edition brings you a curated selection of impactful news, insightful analyses, and exciting advancements from the dynamic world of generative AI. Stay tuned for a concise and informative exploration of this rapidly evolving field.

1. Vidu AI: Revolutionising Video Creation

Vidu AI, a groundbreaking text-to-video AI model from China, has been developed by Shengshu Technology and Tsinghua University. The model uses advanced U-ViT technology to create high-definition videos from simple text prompts. It features multi-camera transitions and can generate realistic or fantastical visuals, making it suitable for a range of creative projects.

While it excels at producing short clips, its potential extends to storyboarding and concept videos, positioning it as a formidable competitor in the AI video creation market. Read more

2. Google's Gecko Benchmark Sets New Standard in AI Image Generation

Google's DeepMind has introduced Gecko, a new benchmark specifically designed to assess AI text-to-image models. Gecko utilizes a structured approach, dividing evaluation into various skills and sub-skills, and introduces a QA-based metric that aligns closely with human judgment. This methodology allows for a nuanced comparison of models, offering insights into each model's strengths and weaknesses in image generation. Notably, Google's Muse model has outperformed competitors like Stable Diffusion on the Gecko benchmark. Read more

3. The Brief Rise and Disappearance of the Mysterious GPT2-Chatbot

A mysterious new AI chatbot named "gpt2-chatbot" briefly surfaced on the LMSYS Org language model benchmarking site, showcasing impressive capabilities comparable to advanced AI models like GPT-4. Despite high traffic and intense public interest, the chatbot disappeared shortly after its debut, leaving behind speculation about its origins and capabilities. Discussions suggest it might be an experimental model from a major AI developer, with LMSYS hinting at a possible future release. Read more

4. Vibe-Eval: New Benchmark for Multimodal AI Evaluation

Reka AI has launched Vibe-Eval, a new benchmark suite designed for evaluating multimodal language models. This suite features 269 high-quality image-text prompts designed to challenge even the most advanced models.

Vibe-Eval aims to differentiate model capabilities clearly and includes a lightweight automatic evaluation protocol using Reka Core, which aligns closely with human judgment. The suite is part of Reka's broader efforts to advance the field through rigorous and meaningful assessments. Read more

?? Foundation Model of the Week - Riffusion

Riffusion is a library for real-time music and audio generation with stable diffusion. It is a latent text-to-image diffusion model capable of generating spectrogram images given any text input. These spectrograms can be converted into audio clips.

Try it on Katonic Playground: riffusion

Subscribe for more exciting AI updates in the future. Have a great weekend! ?