🥟 Chao-Down #305 Meta's image generator has a hard time depicting interracial couples, Big tech launches study on AI's impact on jobs, YouTube says OpenAI and others can't train AI on their videos

It seems like Meta’s image creator is in some hot water as it can’t generate people from mixed racial backgrounds.

The Verge first highlighted that Meta’s Imagine AI system regularly failed to generate images of White and Asian couples together, even when users prompted it to do so. Other channels like CNN also got similar results when they tried to prompt Imagine to create an image of a Black woman with a White husband, which failed.

Many image generators including Meta’s now ship with the label that its images “may be inaccurate or inappropriate.” Even Google publicly acknowledges that there are underlying issues in how they train such systems as evidenced by their very viral snafu of Gemini a couple months ago.

Are we not ready for large scale image generation? Or should these generators just refrain from creating pictures of people?

-Alex, your resident Chaos Coordinator.

What happened in AI? 📰

Washington state judge blocks use of AI-enhanced video as evidence in possible first-of-its-kind ruling (NBC News)

Microsoft, Google join study to see if AI will affect jobs (The Washington Post)

YouTube Says OpenAI Training Sora With Its Videos Would Break Rules (Bloomberg)

The AI Industry Is Steaming Toward A Legal Iceberg (WSJ)

The cost of training AI could soon become too much to bear (Fortune)

OpenAI expands its custom model training program (TechCrunch)

Always be Learnin’ 📕 📖

Supporting Diverse ML Systems : Netflix Tech Blog | Netflix TechBlog

A Visual Guide to Mamba and State Space Models (maartengrootendorst.com)

Developers with AI assistants need to follow the pair programming model (Stack Overflow)

Projects to Keep an Eye On 🛠

impredicative/podgenai: GPT-4 informational audiobook/podcast mp3 generator (Github)

plandex-ai/plandex: An AI coding engine for complex tasks (Github)

infiniflow/ragflow: RAGFlow is an open-source RAG (Retrieval-Augmented Generation) engine based on deep document understanding. (Github)

The Latest in AI Research 💡

ChatGLM-Math: Improving Math Problem-Solving in Large Language Models with a Self-Critique Pipeline (arxiv)

Advancing LLM Reasoning Generalists with Preference Trees (arxiv)

Long-context LLMs Struggle with Long In-context Learning (arxiv)

The World Outside of AI 🌎

Seasonal Allergies Are Coming for Us All (The Atlantic)

Exclusive: Tesla scraps low-cost car plans amid fierce Chinese EV competition (Reuters)

How To Recognize Health Misinformation on Social Media (The New York Times)

Why Our Attention Spans Seem to Be Getting Shorter (TIME)

MrBeast calls for slowing down video editing styles (The Washington Post)

The Case Against Crowdfunding in the Work Chat *Bloomberg)

One Last Bite 😋