AI Roundup 10/20 -> 10/27/2022

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

1. A newly released tool called CLIP Interrogator is using OpenAI's CLIP to extract text prompts from existing images (Hugging Face)

2. AI Twitter learns the same tool is also good at roasting people when uploading a selfie (Futurism)

3. Google AI presents a new method to dramatically improve the performance of large language models called Flan-T5, also publishes its checkpoints (Paper)

4. Adobe introduces generative AI to its suite, allowing creators to generate images and explore creative avenues (Analytics India Magazine)

5. Amazon releases MBXP - a code completion benchmark in 10+ programming languages built by using a framework to convert existing evaluation datasets from one programming language to multiple others (Paper)

6. RunwayML releases Stable Diffusion v1.5 which includes new Inpainting capabilities, followed by a takedown request from Stability AI citing IP leak (Hugging Face)

7. Shutterstock plans to integrate DALL-E 2 into its platform, while compensating creators whose images were used for training purposes (The Verge)

8. Google AI announces U-PaLM—a method that substantially improves existing language models while achieving 2x computational savings (Paper)

9. Dolby proposes DAG - a diffusion-based generative model for general audio synthesis, achieving up to 65% relative improvement compared to the existing state-of-the-art models (Paper's website)

10. Adobe presents Motion Mix - an AI motion-based generation tool that converts a person's still image into an animated dancer (The Verge)

Last week’s AI Tidbits