AIR#94 - June 06, 2024
Good morning, AI aficionados! Grab your coffee and get ready to dive into today's edition of AIR: The AI Recon. Leading the headlines is Nvidia's jaw-dropping achievement of hitting a $3 trillion market cap, driven by the surging demand for AI chips. This milestone not only cements Nvidia's place as a titan in the tech world but also highlights the explosive growth and investment pouring into AI technology. It's a clear sign that the AI boom is far from over and is reshaping the landscape of tech giants.
In other exciting news, Microsoft's Aurora AI is making waves by predicting global air pollution in under a minute. This superfast AI model is set to revolutionize atmospheric science with its rapid and accurate forecasts, offering a glimpse into how AI can tackle some of the world's most pressing environmental challenges. Meanwhile, in the realm of gaming, a rather quirky bug has been discovered that makes machine learning 40% worse at NetHack during a full moon. This amusing yet insightful story underscores the importance of accounting for all variables in AI training data.
But it's not all just breakthroughs and bugs; there's also a fascinating debate brewing over the reliability of large language models (LLMs). A new study introduces an info-theoretic metric to detect unreliable LLM outputs and hallucinations, urging a reassessment of these AI systems' trustworthiness. Whether you're here for the big financial news, groundbreaking scientific advancements, or thought-provoking research, today's edition is packed with stories that challenge and intrigue. So, sit back, sip your coffee, and let's delve into the dynamic world of artificial intelligence together!
Business
Nvidia Hits $3 Trillion Market Cap Amid AI Boom
Nvidia hits $3T market cap, driven by AI chip demand, surpassing Apple to become the second-largest public company. π
Waymo Expands Largest Autonomous Ride-Hail Territory in US
Waymo expands its autonomous ride-hail service in Metro Phoenix by 90 sq miles, now covering 315 sq miles, including Scottsdale and tribal lands.
Musk Redirects 12K H100 GPUs from Tesla to xAI
Elon Musk redirected 12K Nvidia H100 GPUs from Tesla to xAI, prioritizing his AI ventures over the automaker's needs.
EndType: AI-Powered Data Extraction from Images, Videos, and PDFs
Extract data from images, videos, and PDFs using AI. Summarize docs, categorize images, and handle bulk uploads effortlessly!
Artificial Intelligence Runs for Mayor in Cheyenne
AI candidate VIC runs for mayor in Cheyenne, aiming for data-driven governance, but legal hurdles loom as AI isn't a registered voter.
Humane Urges AI Pin Owners to Stop Using Charging Case Immediately
Humane warns AI Pin owners to stop using its charging case immediately due to a fire risk from defective battery cells.
Title: GenAI Empowers Rural Communities to Understand Their Health
AI tool MDandMe empowers rural communities by providing personalized health insights, bridging gaps in traditional healthcare access.
Apple's Unlikely Partnership with Sam Altman to Boost AI
Apple partners with OpenAI to boost its AI capabilities, set to announce at WWDC 2024.
Did OpenAI Build the Best Weapon Detection Software with ChatGPT-4o?
OpenAI's ChatGPT-4o sets a new standard in weapon detection, outperforming existing systems by accurately understanding image context.
Engineering
π₯ [GitHub] Autonomous Foosball-Playing Robot by misprit7
Autonomous foosball robot by misprit7 on GitHub plays, defends, and scores! Check it out: www.youtube.com/watch?v=xrwXZXGiP1w
What Kind of Bug Makes Machine Learning 40% Worse at NetHack?
A full moon confused a NetHack AI, making it score 40% lower due to unaccounted variables in its training data. π
Raspberry Pi AI Kit with Hailo-8L NPU
Raspberry Pi's new AI Kit with Hailo-8L NPU boosts AI performance on Raspberry Pi 5, enabling advanced camera demos and neural network acceleration.
[Github] GLM-4-9B: Open-Source Model Outperforms Llama-3-8B
GLM-4-9B, an open-source model by Zhipu AI, surpasses Llama-3-8B in performance across multiple tasks and languages.
[GitHub] Writer Framework: No-Code Frontend, Python Backend for AI Apps
Create AI apps with Writer Framework: no-code frontend, Python backend. Open-source, fast, flexible, and developer-friendly. π
55 TOPS Raspberry Pi AI PC β 4 TPUs, 2 NPUs
Raspberry Pi AI PC with 55 TOPS, 4 TPUs, and 2 NPUs built using various AI accelerators. Complex setup but functional.
Meta's Trillion-Parameter Generative Recommender System
Meta's new generative recommender system scales to 1.5 trillion parameters, boosting recommendation performance by 12.4%. A new era begins!
Editing Files at 1000 Tokens/s with Llama-70B
Llama-70B achieves 13x faster full-file code edits at 1000 tokens/s, outperforming GPT-4 with speculative-decoding for near-instant accuracy.
NPC-Playground: Interact with AI-Powered NPCs in 3D
Explore NPC-Playground, a 3D demo where you interact with AI-powered NPCs, teach them skills, and shape the future of gaming! π
Song of GPT-4o: Created with HeyMusic's Top AI Music Generator
"Song of GPT-4o" created with HeyMusic's top AI music generator. Try it now at HeyMusic.AI! πΉπΊπΈ
Academic
Superfast Microsoft AI Predicts Global Air Pollution in Under a Minute
Microsoft's Aurora AI predicts global air pollution in under a minute, revolutionizing atmospheric science with rapid, accurate forecasts.
[Paper] To Believe or Not to Believe Your LLM
New study introduces an info-theoretic metric to detect unreliable LLM outputs and hallucinations by quantifying epistemic uncertainty.
π₯ [Paper] Alice in Wonderland: Simple Tasks Show Complete Reasoning Breakdown in State-of-the-Art LLMs
State-of-the-art LLMs fail simple reasoning tasks, showing overconfidence in wrong answers and nonsensical explanations. Urgent reassessment needed.
New AI Detects Rare Epileptic Seizures: USC Viterbi Breakthrough
USC Viterbi's new AI detects rare epileptic seizures by analyzing brain interactions, improving diagnosis with limited data.
Rethink AI from Electrons to Algorithms, Says Stanford Prof
Stanford prof urges a rethink of AI tech stack from electrons to algorithms to address energy crisis, emphasizing biological models over digital.
AI Outperforms Humans in Theory of Mind Tests
AI surpasses humans in Theory of Mind tests, mimicking understanding of mental states, sparking debate on AI's true cognitive abilities.