TLDR AI 2023-09-04

Founded by leaders in the retrieval research domain, this group hopes to allow enterprises to build on top of generative technologies by cheaply and effectively using their own data.

Ex-Meta researchers founded EvolutionaryScale, raising $40M and valued at $200M to fast-track AI-driven protein structure predictions. With claims of outpacing Google's AlphaFold in speed, the startup eyes breakthroughs in medicine and biotech.

The first public work on scaling between these two efforts found that they yield essentially identical final model performance with humans preferring them ~70% of the time over baselines.

Audio deep learning has seen a revival recently. By using a classifier followed by effect-specific modules, a general system can be built to gradually remove effects such as reverb or delay from already mixed audio.

Another exciting result in the recent string of 3D generation methods. This time the resulting meshes are globally consistent and semantically coherent. There is also room to make this technique much faster.

If you optimize Gaussian splatting over multiple independent 2D viewpoints (e.g. unique famous artworks) you get a fabulous stereographic effect.

An easy-to-use modular API to design and train custom Diffusion Models with PyTorch.

A simple framework for building AI chatbots.

WASM, or web assembly, is a compiler target built into modern browsers. It runs almost as fast as native code and has seen a flurry of interest. This post walks through how to build a plugin system for generative apps using WASM.

This article discusses data around OpenAI and ChatGPT traffic. OpenAI traffic has declined around 29% since May. Around 29% of ChatGPT prompts are related to programming, while education accounted for around 10% of all use cases. Other use cases include writing assistance, content creation, and creative ideation. Charts showing data on how people use ChatGPT are available.

AskMore uses AI to conduct user interviews that run without you.

Easily create videos in minutes with your own on-demand digital actors.

X plans to use the information it collects and other publicly available information to help train its machine learning and AI models.