AIR#39 - April 12, 2024
Good morning, AI adventurers! As you take that first, invigorating gulp of your morning brew, today's edition of AIR: The AI Recon is packed to the brim with tales of breakthroughs, ethical conundrums, and technological marvels that promise to redefine the very fabric of our digital existence. At the heart of todayβs buzz, OpenAI's GPTBot has been caught in a digital feast, gobbling up content with a staggering 3 million page fetches, hinting at the insatiable appetite of GPT-5's training mechanisms. This story isn't just about numbers; it's a glimpse into the colossal scale at which AI models operate, reshaping our understanding of data consumption and the challenges it poses.
In a world where the boundaries of technology and ethics increasingly blur, Tsinghua University's quantum leap in cryptography stands as a testament to the relentless pursuit of innovation. Their breakthrough in polynomial time quantum algorithms for lattice problems isn't just a win for cryptography; it's a giant leap forward that could redefine the security landscape in the quantum computing era. Meanwhile, the scholarly realm is abuzz with the revelations of increased vulnerabilities in LLMs due to fine-tuning and quantization, a stark reminder of the fragility that lurks beneath the surface of these computational titans. These stories are not merely academic; they are a clarion call for robust, external guardrails to ensure the security and integrity of AI systems.
As you sip on your coffee and scroll through today's edition, let the stories of AI-driven transformations, from the realms of quantum computing to the intricate challenges of AI security, ignite your imagination. The landscape of artificial intelligence is a tapestry of human ingenuity, ethical quandaries, and technological marvels, each thread interwoven with the potential to reshape our future. So, let's dive into the digital unknown, where each story is a doorway to the wonders and warnings of the AI odyssey.
Business
Apple's M4 Chips with AI Capabilities to Debut in Macs by Late 2024
Apple's late 2024 Macs will feature M4 chips, enhancing AI performance across the lineup, with up to 512GB memory on a more efficient 3nm process.
Huawei Unveils First AI-Powered PC with Intel Chip Amid US Clampdown
Huawei launches AI PC with Intel chip, defying US restrictions. HarmonyOS and Pangu LLM power the innovation.
Adobe Buys Videos at $3/Min to Fuel AI Text-to-Video Generator Amid OpenAI Rivalry
Adobe pays $3/min for videos to fuel its AI text-to-video generator, racing to match OpenAI's tech advancements.
Humane AI Pin Review: A Futuristic Flop
The Humane AI Pin, a device promising a post-smartphone future, falls short with poor functionality and reliability despite ambitious goals. #TechFlop
Apple to Revolutionize Macs with AI-Driven M4 Chips
Apple gears up to revamp Macs with AI-enhanced M4 chips, promising a future of smarter computing.
Introducing Rerank 3: Revolutionizing Enterprise Search and Retrieval
Rerank 3 revolutionizes enterprise search with improved performance, multilingual support, and efficient retrieval, transforming data access for businesses globally.
Air Force Secretary to Test AI-Piloted F-16 in VENOM Program Demonstration
Air Force Secretary to fly in AI-piloted F-16, showcasing confidence in VENOM program's autonomous tech with human oversight.
Intel's Gaudi 3 AI Chip Challenges Nvidia's H100 Dominance
Intel's Gaudi 3 AI chip may outperform Nvidia's H100, promising 50% faster AI training and inference speeds, with a focus on power efficiency and cost.
OpenAI Fires Researchers for Leaking, Including Sutskever Ally
OpenAI fires two researchers, including a Sutskever ally, for leaking info amid internal tensions and a shakeup post-Altman investigation.
Engineering
π₯ OpenAI's GPTBot Overwhelms Content Farm with 3 Million Page Fetches
OpenAI's GPTBot hits content farm with 3M fetches, hinting at GPT-5's training diet. Need OpenAI contacts to address the spider issue!
[Paper] Holodeck: Automated 3D Environment Creation with Language Prompts
Holodeck transforms text into detailed 3D environments, from Star Wars offices to cat-friendly homes, revolutionizing AI training spaces.
π₯ [Github] Run Hugging Face Transformers in Your Browser with transformers.js
Run Hugging Face Transformers in your browser without a server using transformers.js - bringing state-of-the-art ML to the web easily.
[Github] Hugging Face's Parler-TTS: Open-Source High-Quality Text-to-Speech
Hugging Face launches Parler-TTS: a game-changing open-source TTS model, promising high-quality, natural speech from text.
[Github] Giskard: Open-Source Framework for Evaluating LLMs & ML Models
Giskard, a new open-source framework on GitHub, revolutionizes LLM & ML model evaluation, targeting performance, bias, and security issues.
[Github] rgx.tools: GPT-Powered Regex Generator from Text by Amir Mendelson
Transform text into regex effortlessly with rgx.tools by Amir Mendelson - AI does the regex magic so you don't have to!
[GitHub] ChatBotKit's AI Form Builder with Generative UI in Next.js
ChatBotKit launches AI form builder for Next.js, revolutionizing UI in chatbot conversations.
Academic
[Paper] Increased Vulnerabilities in LLMs Due to Fine-Tuning and Quantization
Fine-tuning and quantization increase vulnerabilities in LLMs, reducing resistance to attacks. External guardrails recommended for security.
π₯ [Paper] Tsinghua University's Breakthrough: Polynomial Time Quantum Algorithms for Lattice Problems
Tsinghua University unveils a quantum breakthrough, solving key lattice problems in polynomial time, a giant leap for cryptography.
Taming the Monkey Mind: My Journey with an AI Coach
Conquered my monkey mind with an AI coach, turning walks into therapy sessions. Voice chats with "Tara" offer 24/7 support, reflection, and growth. #AItherapy
[Github] Stanford-Oval/Storm: LLM Generates Wiki Articles with Research and Citations
Stanford's STORM LLM crafts Wiki-like articles by researching online and citing sources, revolutionizing pre-writing for editors.
Meta's AI Chief Yann LeCun: LLMs Far from Achieving Human-Level Intelligence
Meta's AI Chief, Yann LeCun, debunks the near-future arrival of human-level AI, citing LLMs' deep flaws and advocating for an "objective-driven" approach.
Gary Marcus: Why Scaling Alone Won't Achieve AGI
Gary Marcus warns: Scaling won't lead to AGI, highlighting the need for new approaches to transcend data limitations and achieve true AI reasoning.
AI Porn: Revolutionizing Adult Content and Sparking Ethical Debates
AI-generated porn revolutionizes adult content but sparks ethical debates over consent, impact on sex workers, and potential for abuse.
[Paper] Efficient Infinite Context Transformers with Infini-attention by Tsendsuren Munkhdalai et al.
New AI breakthrough: Infinite Context Transformers with Infini-attention enable LLMs to process endlessly long inputs, revolutionizing language modeling.
[Paper] GoEX: Enabling Autonomous LLMs with Minimal Human Supervision
GoEX revolutionizes LLMs with an undo feature for safer, minimal-supervision real-world actions, opening new autonomous application horizons.