AI Roundup 12/08 -> 12/15/2022

Last week's most interesting AI news in <2 minutes

1. Researchers from Meta AI and UT at Austin introduce LaViLa - a new approach to learning video-language representations by leveraging LLMs (Project website)

2. Researchers from CMU and Google introduce MAGVIT to tackle various video synthesis tasks with a single model (Project website)

3. Deepmind introduces Dramatron, a new AI co-writer assistant catered to theatre and film scripts (Twitter)

4. A group of researchers from the Shanghai AI Laboratory and SenseTime present 3DHumanGAN, a 3D-aware GAN for human image generation with consistent appearance under different view-angles and body-poses (Project website)

5. Google Research and UCLA propose REVEAL - a retrieval visual language model achieving state-of-the-art results on visual question answering and image captioning (Paper)

6. Meta releases data2vec 2.0, a new multimodal algorithm that outperforms its predecessor data2vec 1.0 while being 16x faster, as well as its underlying code and pretrained models (Meta AI)

7. Google researchers present RT-1, a multi-task transformer-based model that tokenizes robot inputs and outputs to enable efficient real-time control (Google AI)

8. Researchers from Snap and the University of Illinois introduce SDFusion - a multimodal AI framework that generates 3D assets from text, images, and 3D-shaped inputs (Paper)

9. Google researchers publish their work on Speaker Labels, a new speaker diarization method combining three ML models for its native Recorder app (Google AI)

10. Researchers from Adobe and Carnegie Mellon present SmartBrush - a new model that generates high-quality object inpainting results by using both text and shape guidance (Paper)

11. A group of researchers propose AV-CAT - a Audio-Visual Context-Aware Transformer framework, which produces accurate lip-sync with photo-realistic quality by predicting the masked mouth shapes (Project website)

12. Microsoft introduces Structured Prompting - a new approach to in-context learning in LLMs that breaks in-context learning's length limit and scales to thousands of examples (Paper)

Last week’s AI Tidbits

---

As AI activity slows down towards the end of the year, we will resume our weekly AI Tidbits issues in January. Happy holidays! 🎄