Adam Coates has been tweeting about a variety of topics lately. He has shared materials from the Deep Unsupervised Learning course at Berkeley, covering topics such as AR/PixelCNN, flow models, VAE, GAN, and self-supervised learning. He has also posted a blog post on training neural networks and shared slides for his model-based RL tutorial.

In terms of new trends in AI, Adam Coates has mentioned the use of bigger models for better results and the development of non-autoregressive generative models for speech synthesis. He has also highlighted the importance of diversity in the field and managing competing priorities in data projects.

Overall, Adam Coates' sentiment about AI seems positive. He is enthusiastic about the advancements in deep learning and its ability to make great predictions. However, he is also realistic about the challenges and failures that come with working in the field.