Key AI developments for the fourth week of July 2024, covering LLM competition, AI applications, and data ethics: (1) LLM market — Meta released "Llama 3.1" open-source model with 405B parameters, competitive with GPT-4o and Claude 3.5 Sonnet; Mistral AI released "Large 2" (123B parameters) matching Llama 3.1 performance; OpenAI providing free fine-tuning for "GPT-4o mini"; Microsoft supporting serverless fine-tuning for "Phi-3" on Azure; NVIDIA offering custom LLM development through "AI Foundry" service in collaboration with Meta's Llama 3.1; The Information projected OpenAI may face up to 7 trillion won losses this year despite CEO Sam Altman emphasizing AGI development over profitability. (2) AI applications — Stability AI released "Stable Video 4D" generating multi-viewpoint video from single 3D video clips for film, gaming, and AR/VR applications; OpenAI announced "Rule-Based Reward (RBR)" for safer AI training with consistent safety policy application; Google DeepMind released "AlphaProof" and "AlphaGeometry2" achieving high success rates on International Mathematical Olympiad problems; Las Vegas became first US city to deploy AI gun detection across its entire public transit system, raising both safety and privacy concerns. (3) Data ethics — Runway accused of using YouTube videos without authorization for "Gen-3 Alpha" training, including channels from Korean entertainment companies JYP, SM, YG, and HYBE; X (Twitter) defaulting to allowing user posts for AI training with opt-out option per EU GDPR, criticized for limiting user rights. Key insight: LLM open-source advancement lowers AI access barriers, but data ethics, responsible AI development, and regulatory compliance remain critical unresolved challenges.