AIR#124 - July 09, 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 SwiftA, a lightning-fast, open-source voice assistant powered by Groq, Cartesia, VAD, and Vercel. This new tool promises to revolutionize the way we interact with technology, making voice commands more efficient and accessible than ever before. If you're a developer eager to explore cutting-edge AI-driven creativity, this is a must-read.
But that's not all! Google's latest AI training technology, JEST, is making waves with its claim of being 13 times faster and 10 times more energy-efficient. This breakthrough promises to set new standards for AI training, offering massive gains in speed and energy savings. Meanwhile, Meta AI has unveiled MobileLLM, a compact and efficient language model designed specifically for smartphones. This development challenges the need for massive AI models and could pave the way for smarter, more efficient mobile applications.
And for a touch of the uncanny, Bitcoin mining companies are pivoting to AI to stabilize their revenue streams amid crypto volatility. By leveraging their existing infrastructure, these miners are finding new ways to stay profitable and investors couldn't be happier. 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
Why Bitcoin Mining Companies Are Pivoting to AI
Bitcoin miners pivot to AI for stable revenue, leveraging existing infrastructure amid crypto volatility. Investors are thrilled.
SF's AI Boom Can't Halt Record Office Vacancies
SF's AI boom can't stop record office vacancies; hybrid work and tech layoffs push vacancy rates to 34.5%, rents lowest since 2015.
Chinese Self-Driving Cars Log 1.8M Miles on U.S. Roads, Raising National Security Concerns
Chinese self-driving cars have logged 1.8M miles on U.S. roads, raising national security concerns over data collection and potential espionage.
NY Times Sues OpenAI While Experimenting with Its Tech
NY Times sues OpenAI for article scraping while secretly testing its AI for headline generation. Balancing legal action and tech adoption.
Octolens: AI-Powered Social Listening for B2B on LinkedIn, Reddit, HN
Octolens: AI tool for B2B social listening on LinkedIn, Reddit, HN.
Cyclists Torn Between Fear and Hope for Self-Driving Cars
Cyclists are divided: will self-driving cars make roads safer or pose new dangers? 🚴♂️🚗
AI Industry Must Earn $600B Annually to Cover Hardware Costs, Warns Sequoia
AI must earn $600B annually to cover hardware costs, warns Sequoia. Fears of an AI bubble grow as investments outpace revenues.
The Future is All Bot vs. Bot: AI Turf Wars Over Data
AI bots are battling each other online for data dominance, with new tools like Cloudflare's blocking AI bots to protect content.
Gurman: Apple’s First AI Home Device Will Be a Table-Top Robot
Apple's first AI home device will be a table-top robot, not an upgraded HomePod. Siri fans, keep waiting!
Nicolas Cage Is 'Terrified' of AI
Nicolas Cage fears AI, worried about posthumous use of his likeness: "What are you going to do with my body and face when I’m dead?"
How Chatbots are Battling Global Phone Scammers
AI chatbots are battling global phone scammers by keeping them on the line, collecting data, and disrupting their operations.
Engineering
SwiftA: Fast, OSS Voice Assistant Powered by Groq, Cartesia, VAD, and Vercel
SwiftA: lightning-fast, open-source voice assistant using Groq, Cartesia, VAD, and Vercel. Discover more now! 🚀
Google's New AI Training Tech: 13x Faster, 10x More Efficient
Google's new JEST AI training tech is 13x faster and 10x more efficient, revolutionizing AI training with improved speed and energy savings.
Meta AI Unveils MobileLLM: Compact Language Model for Smartphones
Meta AI reveals MobileLLM, a compact, efficient language model for smartphones, challenging the need for massive AI models.
[Paper] Learning to (Learn at Test Time): RNNs with Expressive Hidden States
New RNN layers with Test-Time Training (TTT) offer linear complexity and improved performance in long contexts, rivaling Transformers and Mamba.
How to Securely Integrate AI Bots into Your Application
Securely integrate AI bots into your app using fine-grained authorization with Permit.io and Arcjet to protect data and users.
Amazon Rufus AI Fails to Filter Non-Product Questions
Amazon's Rufus AI praised for coding skills but fails to filter non-product queries.
Guaranteed Cheapest LLM Calls with No Vendor Lock-In
Switch LLM providers easily and save up to 33% on token costs with no vendor lock-in. Optimize for speed or cost based on your needs. 🚀
MIT Researchers Unveil GenSQL: Generative AI for Databases
MIT unveils GenSQL, a generative AI tool for easy, accurate statistical analysis of database data with just a few keystrokes.
AI Beats Top Racers at Gran Turismo 7 – No Cheating
AI GT Sophy outpaces top human racers in Gran Turismo 7 using only player-visible data, proving AI's prowess without extra info. 🚗💨
Edge-Device LLM Showdown: NeurIPS 2024 Challenge
NeurIPS 2024 challenges AI experts to optimize LLMs for edge devices, tackling memory, energy, and offline functionality issues.
Alibaba Cloud Reveals Custom Datacenter Design for LLM Training
Alibaba Cloud unveils custom datacenter design with 15,000 GPUs for efficient LLM training, leveraging Ethernet to avoid vendor lock-in.
Using LLM and RAG to Generate Insights from My Blog Posts
Built an insight generator using LLM & RAG to extract themes and insights from blog posts, enhancing idea connections and reflections.
Academic
The Moment We Stopped Understanding AI: AlexNet [Video]
Watch how AlexNet changed our understanding of AI forever in this revealing video.
[Paper] Stress-Testing LLMs: Reasoning or Next Token Prediction?
New benchmark MMLU-SR reveals LLMs struggle with true comprehension, showing significant performance drops when key terms are altered.