AIR#127 - July 12, 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 Felix Reda's bold claim that GitHub Copilot doesn't infringe on copyright. This declaration could be a game-changer for developers who rely on Copilot for coding assistance, alleviating fears about legal repercussions. If you're a developer or tech enthusiast, this story is a must-read to understand the evolving landscape of AI and intellectual property.
But that's not all! Privacy concerns are bubbling up as Gemini's auto-summarization feature in Google Docs starts summarizing private documents without permission. This unsettling development raises significant questions about user consent and data security. Meanwhile, Xiaomi is making headlines with its self-optimizing factory in Beijing, capable of producing over 10 million phones annually without human intervention. This technological marvel promises to revolutionize manufacturing efficiency and productivity.
And for a dash of the uncanny, AI has revealed the 'phonetic alphabet' of sperm whales, uncovering a complex communication system akin to human language. This breakthrough offers a fascinating glimpse into the minds of these ocean giants. 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
Gemini Auto-Summarizes Private Docs in Google Docs Without Permission
Gemini auto-summarizes private Google Docs without permission, raising privacy concerns among users.
Xiaomi's Self-Optimizing Factory to Produce 10M+ Phones Annually
Xiaomi's fully automated factory in Beijing will produce over 10M phones annually, optimizing and fixing itself without human intervention.
OpenAI Unveils 5-Level System to Track Path to Superintelligent AI
OpenAI introduces a 5-level system to track progress toward superintelligent AI, currently at Level 1, nearing Level 2 "Reasoners."
Lattice Makes History with First Digital Worker Integration
Lattice makes history by officially integrating digital workers into their HR system, revolutionizing AI employment.
Tesla Prioritizes Musk's and VIPs' Data for Self-Driving AI
Tesla prioritizes data from Elon Musk and VIPs for self-driving AI, potentially skewing development focus and resources.
Odyssey: Hollywood-Grade Visual AI Revolution
Introducing Odyssey: Hollywood-grade visual AI for storytellers to craft cinematic masterpieces with full creative control. Coming soon! 🎬
iLounge and TUAW Return as AI-Driven Content Farms, Steal Writers' Identities
Shady firm relaunches old tech blogs using AI to fake writers' identities, publishing AI-generated content under their names.
Disinformation from Russian AI Spam Farm Tops Google Search
Russian AI spam farm spreads false story about Ukrainian First Lady buying a Bugatti, topping Google search results.
Engineering
GitHub Copilot: No Copyright Infringement, Says Felix Reda
GitHub Copilot doesn't infringe copyright, says Felix Reda. Copilot uses public code for AI training, and its outputs aren't copyright violations.
🔥 Physics-Based Deep Learning Book by TUM
TUM releases a comprehensive book on Physics-based Deep Learning, featuring hands-on Jupyter notebooks for integrating physical models into AI.
[GitHub] Mandala: Effortless Experiment Tracking for Python Computations
Mandala: Simplifies Python experiment tracking by integrating persistence logic and best practices, eliminating code overhead.
🔥 FlashAttention-3: 2x Faster Attention with Asynchrony and FP8
FlashAttention-3 boosts Transformer speed by 2x with asynchrony and FP8, utilizing up to 75% of H100 GPU's capacity.
🔥 [GitHub] Korvus: Unified RAG Pipeline with Postgres
Korvus: Unifies RAG pipeline in a single Postgres query with Python, JavaScript, Rust, and C bindings for high-performance search.
🔥 If AI Chatbots Are the Future, I Hate It
Frustrated with AI chatbots mistaking WiFi issues for Internet problems, Jeff Geerling longs for better human tech support.
🔥 [GitHub] Karpathy: Reproduce GPT-2 (1.6B) in 24h for $672 with llm.c
Reproduce GPT-2 (1.6B) in 24 hours for $672 using llm.c on a single 8XH100 node. No complex libraries needed, just C/CUDA. 🚀
[Paper] Datadog's Toto: State-of-the-Art Time Series Transformer for Observability
Datadog's Toto sets a new benchmark in time series forecasting, excelling in observability metrics with a trillion data points.
GenAI: Brilliant Imitator, Flawed Thinker
GenAI excels in NLP but struggles with logic and true understanding, making it unreliable for tasks needing consistent, logical responses.
Destroying Russian Tanks Is Just the Start for U.S. AI Drone Autopilot
U.S. AI drone autopilot, Skynode S, outperforms human pilots in Ukraine, shrugging off jamming and ensuring higher hit rates on targets.
Toto: Datadog's State-of-the-Art Time Series Forecasting Model
Datadog introduces Toto, a cutting-edge time series forecasting model optimized for observability data, outperforming existing models.
[GitHub] Finetuning DINOv2 for Image Segmentation with LoRA
Finetuning DINOv2 for image segmentation with LoRA simplifies adaptation to new tasks without altering original encoder weights.
Ollama vs. OpenLLM: Concurrency Showdown on Llama 3 8B Model
OpenLLM 0.6 offers fast inference for high concurrency on Llama 3 8B, while Ollama excels in local deployment but struggles with scale.
Academic
Early Apple Tech Bloggers' Work AI-Zombified
Early Apple tech bloggers discover their names and work have been AI-plagiarized on a revived TUAW site, sparking outrage.
The sperm whale 'phonetic alphabet' revealed by AI
AI reveals sperm whale 'phonetic alphabet,' uncovering complex communication akin to human language. A step towards understanding these giants.
Evaluating Chunking Strategies for Retrieval Performance
Chunking strategy choice can boost retrieval performance by up to 9%. Evaluating token-level relevance shows significant impact on accuracy and efficiency.
At least 10% of Research Already Co-Authored by AI
10% of scientific papers are now co-authored by AI, enhancing productivity but raising concerns about bias and misinformation.