AIR#16 - March 20, 2024

Good morning, digital dreamers! Today's AIR: The AI Recon is overflowing with tales of innovation and reflection that are sure to make your morning read an adventure. At the forefront, Nvidia's ambitious Project GR00T is not just a leap but a moonshot towards creating a general-purpose foundation model for humanoid robots. Imagine a future where robots aren't just assistants but partners in progress, capable of achieving AGI in the physical realm. This story isn't just about technology; it's about redefining our relationship with machines and the endless possibilities that await.

Then, there's a moment of nostalgia mixed with awe as Go champion Lee Sae Dol looks back on his historic match against AlphaGo. Eight years have passed, but the lessons and the marvel at AI's evolution remain fresh. It's a reminder of how far we've come and a hopeful glimpse into a future where human and AI collaboration could push the boundaries of what's possible, not just in games but in every aspect of life and learning.

But it's not all rosy; the edition also delves into the critical voices shaping our understanding and approach to AI. Alan Kay's skepticism about teaching programming with LLMs challenges us to think deeper about the systems we trust and the direction of AI education. It's a call to action for a systemic change, ensuring that as we march forward into this brave new world of AI, we're not just carried away by the waves of innovation but are steering the ship with wisdom and foresight. So, buckle up, grab your coffee, and let's dive into the stories that are shaping our digital future.

Business

DeepMind Co-founder Mustafa Suleyman Moves to Microsoft to Spearhead AI Initiatives
Mustafa Suleyman, co-founder of DeepMind, joins Microsoft to lead AI efforts, including Copilot, aiming to innovate consumer AI products.

AWS and NVIDIA Forge Ahead with 414 Exaflops AI Supercomputer for Generative AI Breakthroughs
AWS and NVIDIA team up to create a 414 exaflop AI supercomputer, boosting generative AI capabilities across industries, especially healthcare.

Denmark and NVIDIA to Launch World's Top AI Supercomputer
Denmark & NVIDIA partner to build a top AI supercomputer, aiming to boost research in health, green tech, and more, with a focus on societal benefits.

Teacher Arrested for Creating Child Erotica Using AI and Yearbook Photos
Florida teacher arrested for using AI to create child erotica from yearbook photos.

Inside Google's Woke AI Debacle: Ex-Employees Reveal How Diversity Overtook Excellence
Ex-Google employees claim the company's focus on diversity over excellence led to the flawed AI tool Gemini, sparking a debate on corporate culture vs innovation.

Nvidia and Hippocratic AI Unveil $9/hr AI Nurses, Challenging Healthcare Costs
Nvidia and Hippocratic AI introduce $9/hr AI nurses, offering real-time medical advice via video, potentially revolutionizing healthcare costs.

SXSW AI Praise Met with Audience Boos: A Tech Backlash Brewing
SXSW AI showcase turns sour as audience boos tech praise, signaling a tech backlash is brewing among even the most forward-thinking.

Apple and Google in Talks to Integrate Generative AI into iPhones
Apple and Google may partner to bring generative AI to iPhones, potentially integrating Google's Gemini model to enhance device capabilities.

ByteDance Researcher Mistakenly Added to US AI Safety Groupchat
ByteDance researcher mistakenly joined US AI safety chat, removed for violating code on misrepresentation amid TikTok divestiture debate.

DHS Pilots Generative AI for Immigration Officer Training Amid Bias Concerns
DHS pilots generative AI for training immigration officers, sparking bias and accuracy concerns, despite tech giants' support.

Engineering

Nvidia's Project GR00T: A Moonshot to Create a General-Purpose Foundation Model for Humanoid Robots
Nvidia launches Project GR00T to develop a foundation model for humanoid robots, aiming for AGI in the physical world.

🔥 [Paper] MIT's FeatUp: High-Resolution Revolution in Computer Vision
MIT's FeatUp algorithm revolutionizes computer vision, enhancing detail & accuracy for tasks like object recognition & medical imaging.

[Paper] LLM-FP4: Achieving 4-Bit Floating-Point Quantization in Transformers
LLM-FP4: New 4-bit FP quantization for LLMs cuts resource use, nearly matches full-precision performance on reasoning tasks.

UC Berkeley's RAFT: Elevating LLMs in RAG Performance with Meta Llama 2 and Azure AI
UC Berkeley's RAFT method boosts LLMs in RAG tasks using Meta Llama 2 & Azure AI, merging retrieval with fine-tuning for sharper domain adaptation.

Conversing with Ubisoft's Revolutionary AI NPCs: A Glimpse into the Future of Gaming
Ubisoft's experimental AI NPCs offer a new level of interaction in gaming, blending emotion and decision-making in a creative test.

Early Impressions: GPT-4 Fine-Tuning Outshines GPT-3.5
GPT-4 fine-tuning leaps ahead, outperforming GPT-3.5 by 56% in domain-specific queries, despite higher costs and latency. 🚀

Academic

Go Champion Lee Sae Dol Reflects on Historic AlphaGo Match 8 Years Later
Lee Sae Dol reflects on losing to AlphaGo in 2016, amazed by AI's rapid progress and optimistic about future human-AI collaboration in Go and beyond.

🔥 Alan Kay's Skepticism on Teaching Programming with LLMs
Alan Kay criticizes current programming education, emphasizing trust in complex systems and doubts the efficacy of LLMs in teaching programming, advocating for deeper, systemic change.

[Paper] MindEye2: Transforming fMRI Data into Images with Only 1 Hour of Training
MindEye2 transforms fMRI data into images with just 1hr of training, revolutionizing brain activity visualization.

🔥 Margaret Mead and John von Neumann: Pioneers in the Dawn of AI
Mead and von Neumann, early AI visionaries, pondered digital futures and self-improving systems in the '50s, eclipsing today's AI debates.

🔥 Unveiled: Joseph Weizenbaum's Unpublished Vision for ELIZA's Future
Discovered: Joseph Weizenbaum's draft outlining ELIZA's future in education and conversational AI, hinting at a mathematical model for progress.

AI-Designed Antibodies: A New Frontier in Drug Discovery
AI breakthrough: Scientists design new antibodies from scratch, revolutionizing drug discovery and promising a future where meds are made at the push of a button.

AI Researchers Use AI to Review Peer Work: A Stanford and NEC Labs Study
Stanford & NEC Labs study reveals AI helps review scientific papers, raising questions on authenticity and the need for transparency in AI-assisted peer reviews.

Simple Prompt Skyrockets Claude 3 and Mistral Beyond GPT-4 in API Performance
A simple prompt tweak propels Claude 3 and Mistral's API performance, outshining GPT-4 in innovative benchmark tests.