🥟 Chao-Down #30 A 10x increase in the cost of search, Meta releases LLaMA, Controlling Stable Diffusion models

A new challenger approaches.

Meta just announced their LLaMA large language model. Here’s what you need to know.

LLaMA is smaller than GPT-3, but superior at it in some tasks: LLaMA has >10x fewer parameters (13B vs 175B) and fares well against other models like Google’s PaLM across several benchmarks.

The code is open, but not (yet) the weights: Meta wants LLaMA to be open-source and has put the code online on Github but has yet to release the actual model weights.

Researchers can apply to get access: Meta will allow only a select number of researchers and government organizations access to the model. 

How will LLaMA actually perform against its competitors? We’ll have to wait and find out.

-Alex, your resident Chaos Coordinator.

What happened in AI? 📰

ChatGPT-style search represents a 10x cost increase for Google, Microsoft – (Ars Technica)

Introducing LLaMA: A foundational, 65-billion-parameter large language model (Meta AI Blog)

Qualcomm demos fastest local AI image generation with Stable Diffusion on mobile (The Verge)

Always be Learnin’ 📕 📖

lllyasviel/ControlNet: Let us control diffusion models! (Github)

[Tutorial] Complete Guide to ControlNet for Stable Diffusion img2img (AiTuts)

Projects to Keep an Eye On 🛠

Use AI to create custom MIDI beats in any tempo, key, or mood (Lemonaid Music)

Future of AI in VC (Twitter)

The Latest in AI Research 💡

Fair Diffusion: Instructing Text-to-Image Generation Models on Fairness (arxiv)

Modular Deep Learning (arxiv)

Vid2Avatar: 3D Avatar Reconstruction from Videos in the Wild via Self-supervised Scene Decomposition (moygcc.github.io)

The World Outside of AI 🌎

MLB changes rules to in disruption of itself to keep up with faster-paced sports (MLB)

Americans in Their 30s Are Piling On Debt (WSJ)

For a longer life, afternoon exercise may be best, a large study shows (The Washington Post)

One Last Bite 😋