AI Roundup 10/13 -> 10/20/2022

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

1. Google Research and Tel-Aviv University releases Prompt-to-Prompt - a model to edit images using text prompts (Github repo)

2. Researchers from Cambridge and Amazon AI introduce MINTAKA - a multilingual dataset designed for experimenting with end-to-end question-answering models (Amazon Science)

3. A new few-shot method called SetFi (Sentence Transformer Fine-tuning) outperforms OpenAI’s GPT-3 on text classification tasks (Blog)

4. Google AI releases a novel pre-training method called Unified Language Learner (UL2), along with the checkpoints for its 20B parameter model, that improves the performance of language models across multiple language tasks (Google AI)

5. DeepMind introduces Perception Test—a first-of-its-kind benchmark for evaluating multimodal systems on real-world video, audio, and text data (DeepMind)

6. Hugging Face releases Diffusers v0.5 to quickly generate images on TPUs using Stable Diffusion in Jax (Hugging Face)

7. A team of researchers from Columbia, the University of Pennsylvania, and UC Santa Barbara creates SafeText - the first benchmark dataset to prevent the deployment of unsafe NLP models (Paper)

8. lex.page - a new AI-powered word processor that made the rounds on Twitter (Twitter)

9. Hugging Face releases a Inpainting capability powered by Stable Diffusion to mask or re-fill dedicated areas of an image (Hugging Face)

10. Whisper, GPT-3, and Stable Diffusion were combined to create a digital assistant that generates images based on human instructions (Twitter)

11. Jasper, an AI-powered writing assistant announces a $125M Series A funding round at a $1.5B valuation (TechCrunch)

Last week’s AI Tidbits