TLDR AI 2023-10-19

The AI space is constantly evolving. Are you keeping up?

Discover how to build LLMs, ML, and models faster and smarter at NODES 2023 on October 26 - sponsored by Neo4j. Meet data scientists from around the world as they share their real-world use of graph technology to fight fraud, predict bottlenecks, deliver real-time recommendations, and more.

This 24-hour event features 3 tracks and 90+ sessions, such as:

🧑‍💻New Occupations With OpenAI and Neo4j

🥊Countering Cyberattacks With LLMs

🤖Powering Chatbots With Knowledge Graphs

👉SAVE YOUR SPOT 👈

Can’t wait until October 26? Check out theRoad to NODES hands-on workshops.

P.S. Enter for a chance to win $300 by inviting your friends to NODES! Learn how here.

Adept has released an extremely simple (no image encoder) and very performant (outperforms 13B models) multimodal model. This article outlines how Adept evaluated the model’s performance and some details about the model.

The newest version of Mujoco, a popular physics-based simulator, has been released. It's much faster and far more stable.

OpenAI began working on Arrakis, an AI model designed to run applications more efficiently, soon after the launch of ChatGPT. However, in a rare setback, the project has failed.

The study introduces Zipformer, a revamped transformer for automatic speech recognition (ASR) that's faster, uses less memory, and outperforms the widely-used Conformer.

Instabilities in training at large scales are hard to reproduce for most researchers. This paper shows how to reproduce instabilities in small models and verifies that many commonly reported fixes work just as well on these small models.

Researchers have developed BitNet, a new design for large language models that uses much less energy and memory.

System prompts allow you to change the behavior of ChatGPT-based language models. This repository contains some very well constructed system prompts that could improve usability based on your preferences.

MemGPT is a local chatbot that can read and write to memory. This allows you to have a local assistant that remembers important context about previous conversations.

This repository introduces MonoSKD, a groundbreaking method for detecting 3D objects using only one image that effectively bridges the knowledge gap between LiDAR and regular RGB-based models.

No doubt AI will have a hand in healthcare - it's just a question of how and by how much. This step-by-step guide is ridiculously helpful if you’re starting an AI-for-healthcare company.

Stanford HAI's new Foundation Model Transparency Index reveals that major AI foundation model developers do not provide enough information on their societal impacts. While Meta's Llama 2 scored the highest in transparency at 54%, followed by BloomZ and OpenAI's GPT-4, none achieved particularly high scores, with concerns raised about areas like data sources, societal impact, and potential biases.

Nvidia and Foxconn are collaborating to create "AI factories" designed to supercharge the development of autonomous vehicles and industrial robots.

ChatGPT’s Browse With Bing is now out of beta testing and DALL-E 3 is in beta.

AI autopilot for content creation & marketing.

OpenAI is building an AI image detector that is correct 99% of the time.