AIR#104 - June 16, 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 fascinating shift at OpenAI, where CEO Sam Altman hints at a potential move to a for-profit model. This could reshape the company's future and influence the broader AI landscape, making it a must-read for anyone interested in the business side of tech. Meanwhile, Amazon finds itself in hot water with revelations of secret GitHub scraping to train its AI models, raising significant ethical questions about data privacy and consent.
But that's not all! Tesla's autopilot is making headlines again, this time crashing into a police car in Orange County while the driver was distracted on their cellphone. This incident adds fuel to the ongoing debate about the safety and reliability of autonomous driving technology. On a lighter note, Terence Tao has announced new AI resources for math enthusiasts and a site dedicated to tackling ErdΕs problems, offering a treasure trove of tools for scholars and hobbyists alike.
In the realm of cutting-edge research, a new paper explores the limitations of current language models like GPT-4 in simulating text-based worlds, underscoring the need for further innovation. And if you're concerned about the darker side of AI, don't miss the thought-provoking piece on data-fueled neurotargeting and its potential threat to democracy. Whether you're here for the business insights, tech mishaps, or groundbreaking research, 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
OpenAI CEO Sam Altman Hints at Potential Shift to For-Profit Model
OpenAI may shift to a for-profit model, says CEO Sam Altman, considering a structure change that could impact its nonprofit board control.
Amazon's Secret GitHub Scraping for AI Training
Amazon secretly scrapes GitHub using multiple accounts to train its AI, raising ethical concerns about data privacy and consent.
Tesla on Autopilot Crashes into Police Car in Orange County
Tesla on autopilot crashes into police car in Orange County; driver was on cellphone. No injuries reported.
'Cheapfake' Biden Videos Enrapture Right-Wing Media, Deeply Mislead
'Cheapfake' Biden videos mislead right-wing media, causing confusion and spreading misinformation about his actions at the G-7 summit.
Scale AI: We Hire for Merit, Excellence, and Intelligence
Scale AI formalizes hiring policy: MEI (merit, excellence, intelligence). Focus on talent drives success, ensuring a meritocratic team.
Meta Halts EU AI Training Amid Regulatory Pressure
Meta pauses AI training in EU due to regulatory pressure, delaying Meta AI launch in the region. Privacy advocates cheer, Meta dismayed.
Free Video Generation with DreamMachine AI β Try It Now
Generate high-quality videos from text and images with Dream Machine AI. Fast, efficient, and accessible to all. Try it now!
Swaayatt Robots: Revolutionizing Autonomous Driving in Adversarial Environments
Swaayatt Robots aims to revolutionize autonomous driving with cost-effective, mapless tech, showcasing a 100 KM/H demo on Indian roads.
Engineering
Supercharging My Local LLM Voice Assistant with RAG
Supercharged my local LLM voice assistant with RAG for faster, scalable responses by optimizing context length and using embeddings. π
π₯ Perplexity AI Caught Using Fake User Agent
Perplexity AI caught using fake user agents to bypass robots.txt restrictions, accessing content unethically.
[GitHub] Melody Agents: Generate Songs with AI
Generate songs with AI: choose a topic, genre, and let Melody Agents do the rest! πΆ Available on GitHub.
[GitHub] Open Source AI "Math Notes" Inspired by Apple WWDC
Open-source AI Math Notes app on GitHub lets you draw equations and get instant results, inspired by Apple's WWDC 2024 demo.
GitHub Copilot Chat Vulnerable to Data Exfiltration via Prompt Injection
GitHub Copilot Chat was vulnerable to data exfiltration via prompt injection, allowing attackers to extract sensitive info. Fixed now.
ZeroPoint's Nanosecond Memory Compression Revolutionizes AI Infrastructure
ZeroPoint's nanosecond memory compression tech boosts AI performance by 50%, cutting power use and costs without major system changes.
[GitHub] Open-Source HAL9000 Voice Assistant
Open-source HAL9000 voice assistant: on-device, easy to build, 3D-printable enclosure, no cloud needed. Available on GitHub.
From RAGs to Riches: Making Your AI Chatbot Smarter with Ollama LLM and Open WebUI
Make your AI chatbot smarter with RAG, Ollama LLM, and Open WebUI. Enhance responses using real-time data without retraining.
[GitHub] Alpha Demo of NotePad.ai - Open Source by Dream-Dimension
Dream-Dimension releases NotePad.ai alpha demo on GitHub, an open-source note-taking app with LLM query capabilities.
Academic
[Paper] Can Language Models Simulate Text-Based Worlds?
Current LLMs like GPT-4 struggle as reliable text-based world simulators, highlighting the need for further innovations.
π₯ Two Announcements: AI for Math Resources & ErdΕs Problems by Terence Tao
Terence Tao announces AI resources for math and a new site for ErdΕs problems. Contributions welcome for both projects.
Data-Fueled Neurotargeting: A Threat to Democracy
Data-fueled neurotargeting weaponizes emotional data to manipulate voters, posing a severe threat to democracy's foundations.
[Paper] Converting In-Context Learning to Weights in Linearized-Attention Transformers
Researchers convert in-context learning into model weights for linearized transformers, making it explicit and permanent with bias terms.
Ray Kurzweil's Immortal Soul in the Cloud
Ray Kurzweil predicts merging with AI for immortality, envisioning a future where human consciousness thrives in the cloud.
[Paper] Understanding Hallucinations in Diffusion Models Through Mode Interpolation
Diffusion models' "hallucinations" stem from mode interpolation, creating artifacts outside training data. A metric reduces 95% of these.
Can We Build Conscious Machines?
Can AI be conscious? The debate hinges on whether consciousness requires biological material or if it can arise from silicon and metal.
[Paper] ChatGPT Is Bullshit
ChatGPT outputs convincing text without concern for truth, making it "bullshit" rather than truthful or hallucinated, argues new paper.