October 2023 - AI Tidbits Monthly Roundup

Multimodal AI soars with Adept’s Fuyu, LLaVA 1.5, and Obsidian, new language models deliver unmatched performance at a fraction of the cost, and a host of new techniques to advance robotics.

Welcome to a subscriber-only edition 🔒 of AI Tidbits, where I curate the firehose of AI research papers and tools so you won’t have to.

This is the monthly curated round-up, so if you're pressed for time and can only catch one AI Tidbits edition, this is the one to read—featuring the absolute must-knows.

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

---

Welcome to the October edition of AI Tidbits, where we unravel the latest and greatest in AI. October was filled with innovative breakthroughs and groundbreaking research showcasing the astounding pace of progress in AI.

Leading the charge were open-source multimodal models with the likes of Adept’s Fuyu, LLaVA 1.5, and Obsidian - the world’s smallest multimodal AI. On the open-source front, Hugging Face released Zephyr, a language model beating Anthropic’s Claude 2 on AlpacaEval, and Distil-Whisper - a speech2text model that is 6x faster compared to OpenAI’s Whisper.

Apple joined the generative AI race with a few new papers (Matryoshka, SAM-CLIP) and Google DeepMind was hard at work with new techniques to generate high-quality training data for robotics.

These and many more exciting updates across multimodal AI, video models, and open-source tools are part of this month’s roundup.

Let's dive in!

---

Overview

• Large Language Models

  • Commercial (5 entries)

  • Research (4 entries)

  • Open-source (8 entries)

• ✨ Special feature: Multimodal AI (10 entries)

• Autonomous Agents (4 entries)

• Image and Video (10 entries)

• Robotics (5 entries)

• Cool Tools (5 entries)

• Open-source (6 entries)

Recent AI Tidbits Deep Dives

Revolutionizing document processing with multimodal GPT

The era of AI-powered SMBs

Open-source Generative AI

Large Language Models (LLMs)

Commercial

1. OpenAI's ChatGPT now supports all of its modes in one conversation: Browsing, Advanced Data Analysis, and DALL-E 

2. Phind releases a new 16k context model that beats GPT-4 at coding at GPT-3.5-like speed

3. Perplexity releases two new language models, pplx-7b-chat and pplx-70b-chat, that substantially outperform Llama 2 according to human evaluators

4. Google brings image generation to its Bard chatbot through its text2image model Imagen 

5. Amazon rolls out a suite of AI-powered image generation tools to help advertisers improve their product images

Editing images with DALL-E and GPT-4V

Research

1. DeepMind presents Step-Back Prompting - a two-step abstraction-and-reasoning process resulting in significant performance gains, including a 27% improvement on TimeQA and up to 36% over other prompting methods

2. Nvidia introduces SteerLM - a technique that enables real-time customization of LLMs during inference, showcasing superior performance on benchmarks and broad applicability across gaming, education, and enterprise sectors

3. CMU and Google introduce AutoMix - directing queries to larger LMs based on smaller LMs' output reliability to reduce costs while maintaining performance

4. Researchers release Self-RAG - a framework and models (7B + 13B) boosting LLMs' accuracy and quality by adaptively retrieving relevant information as needed

Step-Back Prompting performance boost across benchmarks

Open-source

1. Hugging Face releases Zephyr - a series of Mistral-based chat models with comparable performance to Anthropic's Claude 2 on AlpacaEval