• Chaos Theory
  • Posts
  • 🥟 Chao-Down #195 OpenAI scraps "Arrakis" model in rare failure, Honda to start a self-driving taxi service in Tokyo, Google says a data-scraping lawsuit would be a "sledgehammer" to generative AI

🥟 Chao-Down #195 OpenAI scraps "Arrakis" model in rare failure, Honda to start a self-driving taxi service in Tokyo, Google says a data-scraping lawsuit would be a "sledgehammer" to generative AI

Plus, Wired spotlights the millions of workers who are training AI models for pennies.

It looks like not everything OpenAI touches turns to gold.

The Information is reporting that OpenAI tried to develop an AI model named after one of the dystopian hellscapes in the sci-fi work “Dune,” but after mixed results, the project failed to take off.

The model, codenamed “Arrakis”, was supposed to be a “sparser” version of their GPT models that in theory would be more efficient and less costly to run without sacrificing too much performance.

Introduction to Dune: Arrakis - Atlantean Seekers

It was also meant to showcase to OpenAI’s partner Microsoft that it could quickly turnaround new next-generation models to justify the $10+ billion investment made into the company. Unfortunately, Arrakis failed and more resources were directed to making GPT-4 more efficient.

Especially with new entrants in open-source and upcoming releases like Gemini from Google, OpenAI faces stiff competition. Will the company continue to innovate?

-Alex, your resident Chaos Coordinator.

What happened in AI? 📰

OpenAI Says New Tool Will Detect Images Made by Dall-E (Bloomberg)

Honda to Start Self-Driving Taxi Service in Tokyo, CEO Says (Bloomberg)

Millions of Workers Are Training AI Models for Pennies (WIRED)

Google says data-scraping lawsuit would take 'sledgehammer' to generative AI (Reuters)

Google's AI stoplight program is now calming traffic in a dozen cities worldwide (engadget.com)

AI chip smuggling into China: Potential paths, quantities, and countermeasures (Institute for AI Policy and Strategy)

YouTube gets new AI-powered ads that let brands target special cultural moments (TechCrunch)

Always be Learnin’ 📕 📖

The Top Metrics for Usage Based SaaS Companies (mostlymetrics.com)

The Greatest Threat to Generative AI is Humans Being Bad at Using it | Max Woolf's Blog (minimaxir)

Prompt engineering for Claude's long context window (Anthropic)

Projects to Keep an Eye On 🛠

openapistack/docs: Full stack typesafe API-first development with OpenAPI. (Github)

Fuyu-8B: A Multimodal Architecture for AI Agents (adept.ai)

EleutherAI/math-lm - Llemma: an open language model for mathematics (Github)

The Latest in AI Research 💡

Jiahao000/MosaicFusion: MosaicFusion: Diffusion Models as Data Augmenters for Large Vocabulary Instance Segmentation (Github)

Reset It and Forget It: Relearning Last-Layer Weights Improves Continual and Transfer Learning (arxiv)

Think before you speak: Training Language Models With Pause Tokens (arxiv)

The World Outside of AI 🌎

How Disney Packed Big Emotion Into a Little Robot - IEEE Spectrum

Amazon launches Project Kuiper satellite internet prototypes (CNBC)

Meta’s planning a cheaper Quest 3 follow-up that ships without controllers - The Verge

Tire Dust Makes Up the Majority of Ocean Microplastics: Study (thedrive.com)

What's Amazon's market share? 35% or 5%? — (Benedict Evans)

YouTube wants to get you watching more news from ‘authoritative sources’ (The Verge)

One Last Bite 😋

The Writers Guild of America secures provisions to protect writers from AI!