AIR#147 - July 31, 2024

Good morning, AI aficionados! As you sip your morning coffee, get ready to dive into today's edition of AIR: The AI Recon. Leading the headlines is a jaw-dropping revelation from Google DeepMind, which spent a whopping $12.9 million on 10,000 LLM training runs to optimize hyperparameters. This eye-popping expense highlights the high costs associated with cutting-edge AI research and the financial dynamics that drive innovation. If you're curious about the economics behind AI advancements, this story is a must-read.

But that's not all! In a budget-friendly twist, researchers have managed to train large-scale T2I diffusion models from scratch for just $1,890, slashing costs by an impressive 118x. This breakthrough could democratize access to powerful AI tools, making it a compelling read for anyone interested in the future of affordable AI. And speaking of innovation, Not Diamond has introduced the ultimate AI model router, which optimizes speed, cost, and performance by selecting the best model for every task. It's a game-changer for developers looking to streamline their workflows.

And for a touch of the uncanny, California's new AI Catastrophe Prevention Plan aims to avert potential AI disasters but has sparked a heated debate on whether it might stifle current research and development. This story delves into the ethical and regulatory challenges of managing AI's rapid growth. Whether you're here for groundbreaking tech updates, ethical debates, or the latest industry buzz, today's edition is packed with stories that will both intrigue and challenge you. So, sit back, sip your coffee, and let's delve into the dynamic world of artificial intelligence together!

Business

Outlit: AI Agents Revolutionizing Sales Contracts
Outlit's AI agents streamline sales contracts, cutting sales cycles, reducing legal costs, and speeding up approvals. Book a demo now!

Apple Uses Google's Chips, Not Nvidia, for AI Models
Apple used Google's TPUs, not Nvidia's GPUs, to train its new AI models, revealing a significant shift in its AI hardware strategy.

From sci-fi to state law: California's AI Catastrophe Prevention Plan
California's new AI bill aims to prevent AI catastrophes but critics argue it could stifle current AI research and development.

The AI Search War Has Begun
The AI search war is on! Tech giants like Google, OpenAI, and Perplexity race to revolutionize search with generative AI, striking media deals.

White House: No Immediate Need to Restrict Open-Source AI
White House sees no need to restrict open-source AI now, highlighting benefits and ongoing monitoring for potential risks.

Meta to Pay $1.4B to Settle Texas Facial-Recognition Case
Meta to pay $1.4B to Texas for violating privacy with facial recognition tech, settling largest state action to date.

AI Transforms Hurricane and Weather Forecasting - The New York Times
AI revolutionizes weather forecasting, predicting hurricanes like Beryl faster and more accurately, potentially saving lives. 🌪️

SaulLM: The First Open Family of Legal Language Models
Equall releases SaulLM, the first open family of legal language models, outperforming GPT-4 in legal tasks. Now available for free.

Perplexity AI to Share Revenue with Publishers Amid Plagiarism Controversy
Perplexity AI to share revenue with publishers after plagiarism accusations, aiming to enroll 30 publishers by year-end.

Harris and Trump Clash on AI Policy
Harris emphasizes AI safety and regulation, while Trump aims to repeal Biden's AI order, advocating for less regulation and more innovation.

Canva Acquires AI Startup Leonardo.ai to Challenge Adobe
Canva acquires AI startup Leonardo.ai to enhance image generation, intensifying competition with Adobe.

Engineering

Not Diamond: The Ultimate AI Model Router
Not Diamond: The ultimate AI model router, selects the best model for every task, optimizing speed, cost, and performance. Try it now!

[Blog] Summarize YouTube Videos with LLM and yt-dlp by Shekhar Gulati
Summarize YouTube videos with LLMs and yt-dlp! Extract key points from subtitles easily using this new script by Shekhar Gulati.

[GitHub] Redcache_AI – Open-Source Memory Framework for LLMs and Agents
Redcache-AI: Open-source memory framework for LLMs and agents, enabling dynamic applications from chatbots to healthcare. Install via pip.

[Paper] Deep-TEMPEST: Deep Learning to Eavesdrop on HDMI Signals
Deep-TEMPEST uses deep learning to eavesdrop on HDMI signals via unintended electromagnetic emissions, improving text capture by 60%.

Create Your Own Custom AI with AI Studio by Meta
Meta's AI Studio lets anyone create custom AIs for chatting, memes, travel advice, and more—no tech skills needed. Try it now!

[Book] AI-Assisted Data Science by Immanuel Trummer
"AI-Assisted Data Science" by Immanuel Trummer teaches using AI like ChatGPT to streamline data science tasks. Pre-order now via MEAP!

Torchchat: Speeding Up Local LLM Inference on Any Device
Torchchat speeds up local LLM inference for laptops, desktops, and mobiles using PyTorch. Try it at the 2024 PyTorch Conference!

[Github] Assistant-UI: Open-Source AI Chat Component for React
Assistant-UI: Open-source AI chat component for React. Customizable, supports rich content, and gaining traction with devs. Join the community!

Free Access to NVIDIA NIM for Developer Program Members
NVIDIA offers free access to NIM microservices for its 5M Developer Program members, enhancing AI model deployment and development.

AI Takes Over Security for Dev Teams
Cycode AI revolutionizes security for dev teams, enhancing risk mitigation and productivity with advanced AI tools and automated remediation.

Finetuned Llama3 Executes Python & Manipulates Data on PC
Llama3 now runs Python scripts and handles data on your PC, serving as a drop-in replacement for OpenAI in most LLM applications.

Academic

🔥 Calculating the Cost of a Google DeepMind Paper: $12.9M for 10,000 LLM Training Runs
Google DeepMind's latest paper cost $12.9M with 10,000 LLM training runs to optimize hyperparameters, revealing the high expense of AI research.

🔥 [Paper] Stretching Each Dollar: Diffusion Training from Scratch on a Micro-Budget
Low-cost AI breakthrough: Researchers train large-scale T2I diffusion models for just $1,890, drastically cutting costs by 118x.

Do You Think ChatGPT Can Reason? – Machine Learning Street Talk
Can ChatGPT reason? ML Street Talk explores its capabilities and limits in their latest YouTube episode. Tune in for insights!