AIR#154 - August 06, 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 buzz is Meta's unveiling of its RoCE network design, a groundbreaking infrastructure connecting thousands of GPUs to support massive models like LLAMA 3.1 405B. This innovation promises to supercharge AI training at an unprecedented scale, making it a must-read for anyone fascinated by the future of AI capabilities. Meanwhile, Karpathy's nano-llama31 on GitHub is turning heads with its minimal, dependency-free version of Llama 3.1, inspired by nanoGPT. If you're a developer looking for a streamlined approach to training and inference, this one's definitely for you.
But that's not all! In the business realm, Groq has just raised a staggering $640M to scale its AI inference technology, positioning itself as a formidable challenger to Nvidia. With a valuation now hitting $2.8B, Groq's move is a clear signal of the intense competition and rapid innovation in the AI chip market. And speaking of financial dynamics, Google has pulled its controversial "Dear Sydney" AI ad after backlash over concerns about AI replacing human creativity. This story delves into the ethical and societal implications of AI in creative spaces, offering a thought-provoking read for those interested in the intersection of technology and human expression.
And for a touch of the uncanny, Nvidia's daily scraping of a lifetime's worth of videos to train its AI has raised significant legal and ethical questions. This revelation highlights the ongoing debate about data privacy and the ethical boundaries of AI training practices. 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
Groq Raises $640M to Meet Soaring Demand for Fast AI Inference
Groq raises $640M to scale AI inference tech, adding talent and capacity to meet rising developer demand. Valuation hits $2.8B.
AI Chip Startup Groq Raises $640M to Challenge Nvidia
AI chip startup Groq raises $640M to challenge Nvidia, doubling its valuation. Yann LeCun and Stuart Pann join as advisors.
What Do People Ask Chatbots? Mostly Sex and Homework
People mostly ask chatbots about sex and homework, an analysis of 200,000 interactions reveals.
Build a Digital Human with NVIDIA NIM and ACE
Create smart, interactive avatars for customer service with NVIDIA NIM and ACE microservices, but use with caution due to potential risks.
Jeff Bezos' Family Office Bets Big on AI
Jeff Bezos' family office, Bezos Expeditions, is heavily investing in AI, making it their primary focus for 2024.
Elon Musk Sues OpenAI and Sam Altman Again
Elon Musk sues OpenAI and Sam Altman again, accusing them of abandoning their nonprofit mission and engaging in racketeering.
Transcript-Based Video Editing with Reduct
Edit videos easily with Reduct: search, redact, highlight, and collaborate using transcripts. Supports 90+ languages and various formats.
Five US States Urge Musk to Fix AI Chatbot Over Election Misinformation
Five US states urge Musk to fix X's AI chatbot, Grok, over election misinformation ahead of November elections.
Secretaries of State Urge Musk to Fix Grok AI Spreading False Election Info
Secretaries of State urge Musk to fix Grok AI after it spreads false election info, risking voter misinformation in 2024.
Elon Musk: Neuralink and the Future of Humanity | Lex Fridman Podcast #438 [Video]
Elon Musk discusses Neuralink's impact on humanity in Lex Fridman's latest podcast. Watch the full episode on YouTube!
Most AI Startups Are Service Companies Disguised as Product Firms
Most AI startups are actually service companies, not product firms, making them harder to scale and less profitable for VCs.
Elon Musk Revives Lawsuit Against OpenAI and Sam Altman in Federal Court
Elon Musk revives lawsuit against OpenAI, alleging deception by Sam Altman and Microsoft in turning nonprofit into for-profit entities.
OnlyFans Stars Using AI to "Sext" Desperate Simps
OnlyFans stars use AI chatbots to sext fans, reducing workload but struggling with kinks. AI firms help bypass platform's bot ban.
Silicon Valley Parents Enroll Kindergarteners in AI Summer Camps
Silicon Valley parents are enrolling kids as young as 5 in AI summer camps, fueling the tech obsession early. 🧒💻
Engineering
[Paper] RDMA over Ethernet for Distributed AI Training at Meta Scale
Meta unveils its RoCE network design for large-scale AI training, connecting thousands of GPUs to support massive models like LLAMA 3.1 405B.
[GitHub] karpathy/nano-llama31: Minimal Llama 3.1 Implementation Inspired by nanoGPT
Karpathy's nano-llama31 on GitHub: a minimal, dependency-free version of Llama 3.1, inspired by nanoGPT, for easy training and inference.
OpenAI Won’t Watermark ChatGPT Text to Avoid User Backlash
OpenAI won't watermark ChatGPT text to avoid backlash and reduced usage, despite having a highly effective detection system ready.
Leaked Docs: Nvidia Scrapes a Lifetime of Videos Daily to Train AI
Nvidia scrapes a lifetime's worth of videos daily to train its AI, raising legal and ethical concerns.
14TB Drive with Assorted LLM Weights - $229
Get a 14TB drive with assorted LLM weights for $229! Includes top models like Llama3.1, Nemotron-4, and more. Available now at Torrance Computer Supply!
[GitHub] Open-source AI-Driven React Components – Hydra AI by Michael Magan
Open-source Hydra AI generates React components at runtime using AI. Register components, and let Hydra dynamically inject them into your app.
Replacing My Right Hand with AI: How I Coded 3,000 Lines with Claude AI
Broke my hand, coded 3,000 lines with Claude AI. Now I'm hooked on AI coding—faster, smarter, and more efficient. The future is here!
Apple's AI Instructions Revealed: 'Do Not Hallucinate' and More
Apple's AI tools include strict instructions: "Do not hallucinate" and avoid negative themes, ensuring accurate and positive interactions.
The Evolution of Extreme LLM Compression: From QuIP to AQLM with PV-Tuning
Yandex's AQLM + PV-Tuning compresses LLMs to 2 bits, outpacing QuIP. Efficient neural networks now run on standard hardware. 🚀
OpenAI tempers expectations with quieter, GPT-5-less DevDay this fall
OpenAI's DevDay will be quieter this fall, focusing on API updates and developer sessions, with no GPT-5 announcement.
Academic
Google Pulls Controversial "Dear Sydney" AI Ad After Backlash
Google pulls "Dear Sydney" ad after backlash over AI writing a child's fan letter, sparking concerns about AI replacing human creativity.