AIR#97 - June 09, 2024
Good morning, AI enthusiasts! As you sip your morning coffee, get ready to dive into today's edition of AIR: The AI Recon. Leading the headlines is a whistleblower's explosive claim that Amazon violated UK sanctions by selling facial recognition technology to Russia through a shell company. This controversy not only puts Amazon under the microscope but also raises significant concerns about corporate ethics and international regulations in the tech industry.
In other exciting news, Microsoft AI has made a groundbreaking discovery of 18 new battery materials in just two weeks, potentially revolutionizing clean energy storage by significantly reducing lithium use. This remarkable achievement underscores the transformative potential of AI in accelerating scientific discoveries and addressing global challenges like sustainable energy. Meanwhile, the AI community is buzzing with speculation that Apple may finally deliver the Siri we've all been waiting for, with a major AI overhaul set to be unveiled at WWDC 2024. Could this be the moment Siri catches up with its AI counterparts?
But it's not all about corporate drama and technological breakthroughs. A thought-provoking piece warns that AI could hit a data ceiling by 2026, potentially stagnating due to the lack of new training data. This looming challenge highlights the need for innovative solutions to keep AI development on an upward trajectory. Whether you're here for the ethical debates, cutting-edge discoveries, or futuristic tech unveilings, 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
Amazon Violated UK Sanctions by Selling Facial Recognition Tech to Russia, Whistleblower Claims
Whistleblower claims Amazon breached UK sanctions by selling facial recognition tech to Russia via shell company; Amazon denies allegations.
Is Apple Finally Launching the Real Siri?
Apple may finally deliver the Siri we were promised, with a major AI overhaul set to be unveiled at WWDC 2024.
AI's Hidden Workers Trapped in Dead-End Jobs
AI's hidden workers, stuck in low-wage jobs, struggle for economic mobility while training tools for top tech firms.
Insights from the UN's "AI for Good" Summit
UN's AI for Good Summit highlights AI's potential for global goals but raises concerns on transparency, sustainability, and ethical use.
ChatGPT's Grandmother: Siemens' 1974 AI Pioneer
ChatGPTβs roots trace back to Siemens' 1974 AI pioneerβa natural language Q&A system. Discover AI's early history!
The War for AI Talent Intensifies as Big Tech Scrambles
Big tech battles for AI talent as brain drain hits firms like OpenAI, with top researchers leaving amid industry-wide demand.
Engineering
π₯ [GitHub] LSP-AI: Open-Source Language Server for AI Code Assistance by SilasMarvin
LSP-AI: Open-source language server by SilasMarvin enhances software engineers' productivity with AI-powered features across multiple editors.
π₯ Chunking Strategies for Effective RAG Systems
Effective chunking is crucial for RAG systems to ensure accurate, context-rich LLM responses. Smaller, context-aware chunks often yield better results.
AMD's "Peano" LLVM Compiler for Ryzen AI NPUs
AMD's new open-source "Peano" LLVM compiler enhances Ryzen AI NPUs on Linux, boosting AI processing capabilities.
Kling AI: Sora-Like Text-to-Video Generator on Kuaiying App
Kling AI's new Sora-like model on Kuaiying app turns text into stunning 1080p videos, up to 2 mins long, with advanced 3D and cinema-grade quality.
[GitHub] StreamSpeech: All-in-One Model for Offline and Simultaneous Speech Recognition, Translation, and Synthesis
StreamSpeech: An all-in-one model for offline/simultaneous speech recognition, translation, and synthesis. Now on GitHub!
[GitHub] Jockey: Conversational Video Editing Agent by Twelve Labs
Twelve Labs' Jockey: a conversational video editing agent leveraging LLMs and VFMs for seamless video workflows. Now in alpha! π
[GitHub] Peano: LLVM Backend for AMD/Xilinx AI Engine Processors
AMD/Xilinx's Peano: LLVM backend for AI Engine processors is now open source, supporting AIE2 architecture in RyzenAI SoCs.
Introducing Generative Physical AI [Video]
Watch the latest breakthrough: Generative Physical AI in action! ππ #AIInnovation
Microsoft AI Discovers 18 New Battery Materials in Two Weeks
Microsoft AI finds 18 new battery materials in 2 weeks, potentially cutting lithium use by 70%. Revolutionizing clean energy storage.
Academic
AI Hitting Data Ceiling, Could Stagnate by 2026
AI may stagnate by 2026 due to lack of new training data, warns researchers. Scaling models efficiently could become impossible.
[Paper] Finite-Time Topology Identification of Delayed Complex Networks and Its Application
New finite-time method identifies and synchronizes delayed complex network topologies rapidly, with applications in power grid outage detection.
WARC-GPT: AI-Powered Tool for Exploring Web Archives [GitHub]
WARC-GPT: Explore web archives via AI chatbots using Retrieval Augmented Generation. Open-source tool for library pros and researchers.
LLMs Fail at Basic Pangrams: Demian's Blog
LLMs struggle with basic pangrams, revealing their limitations and our overestimation of their "intelligence." Are we hyping AI too much?
AI Is a Hall of Mirrors
AI mirrors our desires, recycling what we know, but fails to offer truly new insights, leaving us yearning for genuine discovery.
Paleontologists Outraged by AI's Inaccurate Prehistoric Animal Depictions
Paleontologists criticize AI for inaccurate depictions of prehistoric animals, arguing it lacks the scientific precision and human touch of traditional paleoart.
[Dataset] Exploring MLP Neurons in Meta's Llama-3-8B Model
Meta releases a dataset of text snippets that activate MLP neurons in Llama-3-8B, aiding transformer model interpretability and research.
[Paper] The Geometry of Concepts in Large Language Models
Study reveals how large language models encode categorical and hierarchical concepts using geometric structures like simplices and polytopes.
[GitHub] Massive Dataset of Jailbreak Prompts for LLMs (15,140 Prompts)
GitHub repo released with 15,140 ChatGPT prompts, including 1,405 jailbreak prompts, for research on LLM vulnerabilities.
HBR: Don't Expect Juniors to Teach Senior Professionals Generative AI
Juniors can't effectively teach senior professionals generative AI due to emerging tech risks and lack of deep understanding, study finds.