AIR#14 - March 18th, 2024
Good morning, AI aficionados! Today's edition of AIR: The AI Recon is brimming with groundbreaking stories that are sure to kickstart your day with a jolt of technological excitement. Leading the charge, Comma.ai's remarkable pivot from steering model failures to pioneering imitation learning in autonomous driving showcases resilience and innovation, proving that the road to AI success is paved with both challenges and breakthroughs. This tale of overcoming the infamous ε error challenge is a testament to the relentless pursuit of perfection in the AI-driven automotive frontier.
In a bold move to democratize AI, Microsoft is unlocking the power of GPT-4-Turbo for Copilot users at no extra cost, marking a significant leap towards enhancing AI capabilities and accessibility for all. This strategic play not only promises to supercharge productivity but also sets a new benchmark in the AI landscape, fostering a more inclusive and innovative environment for developers and businesses alike. Meanwhile, Nvidia's eyes are set on a billion-dollar acquisition of AI platform Run:ai, signaling a robust expansion strategy and underscoring the intense competition and high stakes in the AI industry.
Amidst these titanic shifts, LLM4Decompile is revolutionizing reverse engineering by decompiling binary code with AI, making code readable and re-executable, a boon for developers worldwide. This remarkable innovation, alongside other stellar developments in the AI sphere, underscores the transformative power of AI across diverse domains. So, grab your coffee, and let's dive into these exhilarating stories, exploring the boundless possibilities and the bright future that AI heralds for our world.
Business
Microsoft Unlocks Free Access to GPT-4-Turbo for Copilot Users
Microsoft makes GPT-4-Turbo free for Copilot users, enhancing AI capabilities and accessibility for all.
Nvidia Eyes Billion-Dollar Acquisition of AI Platform Run:ai in First Israeli Deal Since Mellanox
Nvidia in advanced talks to buy AI firm Run:ai for up to $1B, marking its first Israeli acquisition since Mellanox.
Midjourney's Bold Gamble Against Copyright Law Challenges
Midjourney updates its terms, betting big on AI's use of copyrighted content under fair use, despite legal risks and creator disputes.
Hasbro CEO Envisions AI-Driven Future for D&D and MTG
Hasbro CEO sees AI shaping the future of D&D and MTG, despite community concerns and past backlash over AI art use.
Anguilla's AI Domain Boom: A Caribbean Island's Unexpected Fortune
Anguilla's .ai domain boom fuels island's fortune, making tech history as AI companies flock to secure their digital presence.
Engineering
Comma.ai's Journey from Failure to Pioneering Imitation Learning for Autonomous Driving
Comma.ai evolves from failed steering models to pioneering imitation learning in autonomous driving, tackling the ε error challenge.
🔥 [GitHub] LLM4Decompile: Using LLMs for Binary Code Decompilation
LLM4Decompile revolutionizes reverse engineering by decompiling binary code with AI, making code readable and re-executable.
[GitHub] Humanify: Reverse Minified JavaScript with LLMs
Humanify on GitHub uses LLMs to reverse minify JS, turning cryptic code back into readable forms with original variable names intact.
[GitHub] Amazon Science's Chronos: Pretrained Models for Time Series Forecasting
Amazon Science unveils Chronos: groundbreaking AI for precise time series forecasting, leveraging pretrained models.
[Hugging Face] 01.AI Unveils Yi Series: Bilingual Open-Source LLMs Surpassing Llama
01.AI's Yi series revolutionizes LLMs with bilingual, open-source models, outperforming Llama and ranking near GPT-4 in global benchmarks.
[Github] MusicLang: Symbolic Music Generation with LLAMA2 AI
MusicLang unleashes LLAMA2 AI for symbolic music creation, offering control over chord progressions and fast, GPU-free operation. Perfect for DAW exports.
Evaluating Large Language Models: Metrics, Methods, and Innovations
Revolutionizing LLM evaluation: New metrics & methods outshine traditional techniques, promising more accurate & insightful AI assessments.
Comfy AI: Crafting Artistic Backgrounds with Mathematics [Video]
ComfyUI AI transforms math into art, using Fibonacci, Euler, Heisenberg, and Brown's formulas to craft unique backgrounds. Watch on YouTube!
GPT-2 Model Runs in a Massive Excel Sheet: Download from GitHub
GPT-2 now runs in a massive Excel sheet, proving AI's basics can be understood through spreadsheets. Download it from GitHub for a unique demo.
🔥 Neural Networks Revolutionize Image Compression: Insights from Martin Lumiste
Neural networks are revolutionizing image compression, offering better ratios and quality by learning to optimize the trade-off between bitrate and distortion.
Open Release of xAI's Grok-1: A 314 Billion Parameter AI Model
xAI releases Grok-1, a groundbreaking 314B parameter AI model, open-source under Apache 2.0. Dive into AI innovation with Grok-1's architecture now available.
[Paper] Accelerating LLM Inference with Dynamic Memory Compression on NVIDIA GPUs
Dynamic Memory Compression boosts LLM inference speeds up to 3.7x on NVIDIA GPUs without extra parameters, preserving performance.
[Paper] Deep RL Powers Realistic Fruit Fly Locomotion Simulation
Deep RL creates a detailed fruit fly locomotion sim, blending physics and neural networks for realistic behaviors in air and on land.
Academic
The Surprising Future of Work in the AI Era: High Pay and Plenty of Jobs
AI won't end work; it'll create high-paying jobs by making us focus on tasks where humans hold a comparative advantage.
[Paper] How API-Protected LLMs Unintentionally Reveal Proprietary Info
New study reveals API-protected LLMs can leak proprietary info, posing risks and calling for better security measures.
[Paper] Evaluating LLMs in Planning Tasks: A Critical Investigation
New study reveals Large Language Models struggle with autonomous planning tasks, success rate ~12%, but shine as heuristic aids in LLM-Modulo settings, enhancing external planning processes.
[Paper] Google Research's "Talk like a Graph": Improving LLMs with Graph Encoding
Google's "Talk like a Graph" boosts LLMs' graph reasoning by up to 60% with innovative encoding, aiming to revolutionize AI understanding of complex networks.
Wearable Device Enables Speech Without Vocal Folds Through Machine Learning
New wearable device allows speech without vocal folds, using machine learning to convert muscle movements into voice, enhancing life for patients with voice disorders.
[Paper] Enhancing Storm Surge Forecasting with LSTM Machine Learning
LSTM AI boosts storm surge forecast accuracy, training on 61 historical storms for real-time, bias-corrected hurricane predictions.