TLDR AI 2024-07-09

Meta has released pre-trained models that leverage a novel multi-token prediction approach that forecasts multiple future words simultaneously, promising enhanced performance and drastically reduced training times.

DeepMind and Harvard University developed a virtual rat with AI neural networks trained on actual rat movements and neural patterns to probe the brain circuits responsible for complex motor skills. This bio-inspired AI has the capacity to generalize learned movement skills to new environments, offering insights into brain function and advancing robotics. The research demonstrates that digital simulations can effectively mimic and decode neural activity related to different behaviors.

AriGraph is a novel method that helps AI agents build a memory graph integrating both semantic and episodic memories.

PartCraft is a new method in generative visual AI that allows users to select visual concepts by parts, moving beyond traditional text or sketch-based methods.

A post by Neel Nanda, a Research Engineer at Google DeepMind, about his favorite papers to read in Mechanistic Interpretability.

Microsoft's Minference dramatically speeds up supported models' inference with a number of system improvements.

EleutherAI trained SAEs for Llama 3 8B and released weights and the library code.

An AI agent that writes and fixes code for you.

Researchers have developed a new approach to estimating category-level 3D poses from casual, object-centric videos without human supervision.

AnySR is a new approach to single-image super-resolution (SISR) that enhances efficiency and scalability. Unlike traditional methods, AnySR supports the 'Any-Scale, Any-Resource' implementation, reducing resource needs for smaller scales without additional parameters.

PlatoNeRF is an AI technique developed by MIT and Meta researchers that uses shadows and single-photon lidar to create 3D models of scenes, including obstructed areas. This method could enhance the safety of autonomous vehicles and improve AR/VR experiences. PlatoNeRF outperforms traditional methods and has promising practical applications with lower-resolution sensors.

A speech foundation model with multiple speech understanding capabilities, including automatic speech recognition, spoken language identification, speech emotion recognition, and audio event detection.

Doti is an AI-powered assistant for managing health and life goals.

OpenAI's ChatGPT has varied performance in code generation, with success rates ranging from less than 1% to 89% depending on factors like task difficulty and programming language.