AI Roundup 09/07-> 09/14/2023

Stability AI's first generative audio model, DeepMind's magical prompt improving accuracy by 9%, and a single multimodal AI that accepts and generates content across text, image, audio, and video

Welcome to the weekly edition of AI Tidbits, where I curate the firehose of AI research papers and tools so you won’t have to.

Reach AI builders, researchers, and entrepreneurs by partnering with AI Tidbits

---

Highlights

1. Stability AI releases Stable Audio - a text-to-audio generative model capable of creating 95-second music clips (The Verge)

   

2. Researchers introduce NExT-GPT - a system that revolutionizes multimodal LLMs by not only understanding but also generating content in multiple modalities, fostering a path to more human-like AI (Project website)

   

3. Microsoft releases phi-1.5 - a 1.3B Transformer-based model that rivals larger LLMs and exhibits improved behavior due to the exclusion of web data in its training set (Paper)

   

4. Osmo AI builds a model that can predict smells based on molecular structures (Company blog)

   

5. Adept open-sources Persimmon-8B - a commercially permissively licensed LLM outperforming the 7B Llama 2 and MPT models (Adept)

   

6. Researchers release Medusa - a framework that speeds up LLM inference by 2x (Project website)

7. DeepMind proposes a prompt that improves LLMs' accuracy by 9% as part of a new method called OPRO - leveraging LLMs to create prompts, outperforming human-designed prompts in task accuracy (Paper)

   

8. Roblox announces Roblox Assistant - an AI chatbot that helps creators generate virtual environments seamlessly (The Verge)

   

9. Microsoft and MIT partner to create 5k free and open audiobooks using AI (Project website)

10. Deci releases DeciLM 6B - a language model that is 15x faster than Llama 2 7B while maintaining comparable performance (Company blog)

    

11. ByteDance unveils PhotoVerse - a resource-efficient solution for personalized text-to-image generation, capable of creating high-quality images with just a single facial photo (Project website)

    

12. Microsoft pledges legal protection from AI copyright infringement lawsuits for its Copilot and Bing Chat users (Ars Technica)

13. Google releases new free courses to understand and master Generative AI (Google Blog)

Research

1. Humboldt University introduces OpinionGPT - a web demo that transparently showcases LLMs' biases by allowing users to compare responses fine-tuned to different biases: gender, geographics, and more

   

2. University of Waterloo and Ohio State University present MAmmoTH - a series of open-source LLMs specialized for general math problems, outperforming existing ones such as WizardMath and GPT-4

   

3. Salesforce, Columbia, and MIT propose a new prompting technique called Chain of Density (CoD) to improve GPT-4's content summarization capability

4. Researchers unveil DiffBIR - a framework leveraging pretrained text-to-image diffusion models, setting a new benchmark in blind image restoration tasks

   

5. Helixon Research and UT Austin develop InstaFlow - the first one-step diffusion model for rapid text-to-image generation, setting new benchmarks in image quality and generation speed

   

6. Researchers introduce AstroLLaMA - a 7B model fine-tuned from LLaMA-2 using over 300k astronomy abstracts

7. Google open sources MADLAD-400 - a 3T token dataset spanning 419 languages along with a suite of machine translation models (3B, 7B, 10B) (Paper)

8. Cohere finds that a simple perplexity metric is surprisingly effective in pruning pretraining datasets for large language models, retaining performance while using only 30% of the original data

Announcements

1. Slack releases new AI capabilities including instant channel recaps, thread summaries, and workflow automation

2. EY launches EY.ai - an AI platform to help its clients boost AI adoption, along with its own new language model EYQ

3. Intuit announces a host of generative AI features, including Intuit Assist - an AI-powered financial assistant for SMBs

Cool tools

1. AnimateDiff - turn text prompts into short video clips

Open-source

1. Rivet - an interactive development environment for creating complex AI agents

2. ExLlamaV2 - a fast inference library for running LLMs locally on modern consumer-class GPUs

Last week’s AI Tidbits

AI Roundup 08/24-> 09/07/2023

Reach AI builders, researchers, and entrepreneurs by advertising in AI Tidbits

If you find AI Tidbits valuable, share it with a friend and consider showing your support.