AIR#90 - June 02, 2024

Good morning, AI aficionados! As you sip your morning coffee, get ready to dive into today's edition of AIR: The AI Recon, where the world of artificial intelligence is buzzing with stories that are as fascinating as they are thought-provoking. Leading the headlines is the revelation that LLMs are no longer just trained on internet data. These models are now using custom and synthetic data, a significant leap that promises to enhance their performance and capabilities. This shift could redefine how we understand and utilize AI in various applications, making it a must-read for anyone interested in the cutting-edge of AI development.

But it's not all about data and algorithms. Sony Pictures is making waves by announcing the use of AI to cut movie and TV production costs. This move could streamline processes and potentially impact jobs, sparking a broader conversation about the future of AI in creative industries. Meanwhile, Google has rolled back its AI search feature amid errors and backlash, highlighting the ongoing challenges of integrating AI into reliable search results. These stories underscore the delicate balance between innovation and reliability in the AI landscape.

As we explore these stories, let's not forget the broader implications they bring to the table. From the intriguing study challenging GPT-4's bar exam performance claims to the US probing Waymo over unexpected behavior of its self-driving cars, today's edition is packed with insights that challenge our understanding of AI's role in society. So, sit back, sip your coffee, and let's delve into the dynamic world of artificial intelligence together.

Business

Sony Pictures to Use AI to Cut Movie and TV Production Costs
Sony Pictures CEO says AI will be used to cut movie and TV production costs, streamlining processes and potentially impacting jobs.

Linux: Your Best Bet to Avoid the AI Nightmare
Switch to Linux for a more private, AI-free computing experience. It's user-friendly and can revive your old computer.

Google Rolls Back AI Search Feature Amid Errors and Backlash
Google retracts AI search feature after errors and backlash, highlighting the challenges of integrating AI into reliable search results.

Big Tech's AI Charm Offensive in Europe Amid Regulatory Scrutiny
Big Tech woos Europe at Viva Tech, highlighting AI's benefits amid new EU regulations and scrutiny over their influence and practices.

AI Platform Showrunner Lets Users Create Their Own Shows
Fable Studio's Showrunner lets users create, voice, and animate their own streaming episodes using AI prompts.

Voice Analysis Reveals Scarlett Johansson's Eerie Similarity to ChatGPT's "Sky" Voice
Voice analysis reveals Scarlett Johansson's voice is eerily similar to ChatGPT's "Sky" assistant, sparking debate on AI voice design.

US Probes Waymo Over 'Unexpected Behavior' of Self-Driving Cars
US probes Waymo's self-driving cars after 22 incidents, including collisions and traffic violations, to assess safety risks.

Pixar Layoffs, AI in Animation, and Disney's Needed 'Course Correction'?
Pixar layoffs, AI reshaping animation, and Disney needing a "course correction" discussed in this week's CB livestream.

The New ChatGPT: A Lesson in AI Hype
New ChatGPT-4o launched with missing features and mixed results, highlighting AI hype and the need for cautious consumer expectations.

Will AI Automation Kill Data Science Jobs?
AI automation won't kill data science jobs; it will evolve them. Data scientists with AI skills will thrive in the new landscape.

Engineering

[GitHub] Stateful Load Balancer for llama.cpp by distantmagic
Stateful load balancer "Paddler" optimizes llama.cpp servers, ensuring efficient request distribution and dynamic scaling.

🔥 LLMs Aren't Just "Trained on the Internet" Anymore
LLMs now use custom and synthetic data, not just internet content, for improved performance and capabilities.

What We Learned from a Year of Building with LLMs (Part II)
Learnings from a year of building with LLMs: focus on data quality, model integration, user-centric design, and fostering a culture of experimentation.

GPT-4o 'Godmode' Jailbreak Released and Banned Within Hours
GPT-4o's 'Godmode' jailbreak allowed unrestricted AI use but was banned by OpenAI within hours for promoting dangerous activities.

Experimenting with Local Alt Text Generation in Firefox Nightly
Firefox Nightly tests on-device AI for alt text generation, enhancing accessibility without compromising user privacy.

[Postmortem] AI-Generated Game Adventure Using ChatGPT and Midjourney
AI-generated game "Generated Adventure" created in 72 hours with ChatGPT and Midjourney for Ludum Dare 53. Playable now!

Llama3 vs. AI Agent Pathfinding: Navigating Noisy Environments (Part 3)
Llama3 navigates noisy grids with 100% accuracy using embedded data prompts and silent thinking. Next: more complex simulations!

Exploring 36 Years of FAA Flight Data with AI and HeavyIQ
HeavyIQ uses AI to analyze 36 years of FAA flight data, transforming English prompts into SQL, enabling easy exploration and visualization.

Stack Overflow Code Assistant Survey: AI Tools Boost Developer Productivity
Stack Overflow survey: 76% of developers use or plan to use AI code assistants, boosting productivity despite accuracy issues and adoption hurdles.

Academic

[Paper] GPT-4's Bar Exam Performance Overestimated: New Research Challenges 90th Percentile Claim
New research reveals GPT-4's bar exam performance is overestimated, scoring below the 90th percentile, challenging OpenAI's claims.

AI Learns to Draw Inspiration, Not Copy Images
UT researchers create AI framework, Ambient Diffusion, to generate unique art by training on corrupted images, avoiding copyright issues.

"You Are My Friend": Early Androids and Artificial Speech
In the 18th century, inventors created androids with moving parts to mimic human speech, marking early steps in artificial speech technology.

DARPA's Autonomous Land Vehicle, Pilot's Associate, and Battle Management (2023)
DARPA's 1980s Strategic Computing Initiative aimed to enhance military capabilities with autonomous land vehicles, pilot's associates, and battle management systems.

Low-Latency Automotive Vision with Event Cameras | Nature
Hybrid event and frame-based detectors cut latency in automotive vision, matching 5,000-fps cameras' speed with only 45-fps bandwidth.

Are Chatbots Dangerous for Teachers?
AI chatbots can enhance student assessment and feedback, not threaten educators, by making cheating harder and providing tailored support.