AI Roundup 09/15 -> 09/22/2022

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

1. Google publishes new deep learning models for view synthesis using transformers (Google AI)
   

2. OpenAI open source Whisper - a new automatic speech recognition (ASR) system enabling human-level multilingual transcription and translation (OpenAI)

3. Adept showcases a ACT-1, a large-scale Transformer trained to navigate apps and websites such as running a multi-step Zillow query (Adept)

   

4. OpenAI launches a DALL-E new capability to upload and edit photos containing human faces (The Verge)

5. LAION-AI releases a new SOTA image classification OpenCLIP models (LAION)

6. OpenAI experiences malicious prompt injection attacks asking the model to ignore its previous directions and leak the original prompt (Simon Willison’s Weblog)
   

   

7. Google AI introduces ‘PaLI’, a model trained to perform visual question answering, image captioning, and image classification tasks in over 100 languages (Google AI)
   

   

8. A Twitter user shares a comprehensive list of unique applications built using Stable Diffusion (Twitter)
   

   

9. Krea is the new search engine for AI-generated art and prompts (Krea)
   

   

10. Scientists from Harvard and Stanford develop a self-supervised diagnostic tool that can detect diseases on chest X-rays purely without requiring the radiologist's annotated images (Nature)

11. An Israeli AI company launches a new platform allowing users to generate videos using a single image and a textual prompt (TechCrunch)
    

    

12. Scientists teach a Speech Recognition Robot to Laugh (The Batch)

13. NVIDIA releases BioNeMo - a framework for the prediction of protein, molecules, and DNA by training and deploying large biomolecular language models (Analytics India Magazine)

Last week’s AI Tidbits