AI News Summary

Introduction

• In the final days of vacation in Colorado but keeping up with AI news.
• A busy week in AI, featuring major announcements from various companies.

Meta's llama 3.1 Release

• Llama 3.1: New version released; an upgrade from llama 3.
• Model Sizes: Available in three sizes: 8B, 70B, and 405B parameters.
  • Bigger models enhance reasoning, math, and general use cases.
• Features:
  • Tool usage, multilingual capabilities, complex reasoning, better coding skills.
• Benchmark Performance:
  • Outperformed leading models like GP4, Omni, and Claude 3.5 Sonic.
  • Smaller llama 3.1 (8B) surpasses other smaller models in various performance metrics.
• Open Source Advantage:
  • Available for public download and customization, unlike Claude 3.5 Sonic or GPT-4.
  • Licensing caveat for companies with over 700 million active users needing to negotiate with Meta.
• Accessibility:
  • Can interact with models via Meta's AI platforms like WhatsApp and Messenger.
• Grock Partnership:
  • Llama 3.1 models are also integrated on Grock's platform for fast AI inference.
  • Fast performance noted with response times of approximately 250 tokens/second.

Mistral Large 2 Release

• Mistral Large 2: 123 billion parameter model released, competes with llama 3.1.
  • Outperforms Llama 3.1 in math performance and human eval benchmarks.
  • Strong performance in coding languages noted.

Apple's AI Model Introduction

• Released smaller models (7B and 1.4B parameters), claiming to rival earlier versions of llama.

Google Gemini Improvements

• Upgrade to Gemini 1.5 Flash: Improvements in quality and latency, especially in reasoning.
  • Increased token capacity (32k tokens) and file upload capabilities announced.

OpenAI Developments

• Fine-tuning for Free: Users can fine-tune GPT-4 for free up to 2 million tokens per day until September 23.
• Search GPT Prototype: New AI search capability launched to enhance user experience.

Other Notable News

• Anthropic Scraping Concerns: Scraping of websites at an unsustainable rate criticized.
• Grock Updates: Elon Musk announces Grock 3.0 expected to be the most powerful AI model by December.
• Bing Redesign: Bing AI interface redesigned to leverage AI responses more prominently.

AI Video Generation Updates

• Loop Feature by Luma AI: New feature to create infinitely looping animations.
  • Good for turning static images into animated video loops.
• Cing AI Text-to-Video: Impressive text-to-video generation capabilities noted.

Ethical Concerns on Data Scraping

• Runway Training Controversy: Allegations of unauthorized data scraping from YouTube channels.
• AI Fair Use Debate: Exploring the ethical implications of training AI on publicly available data.

Stability AI & Adobe Updates

• Stable Video 4D: Tool to transform single-object videos into multiple angles.
• New AI Features in Adobe Illustrator/Photoshop: Native AI features in design software for creative enhancements.

Sundry AI Tools & Applications

• Sakana AI's Traditional Art Model: New model for generating Japanese Ukiyo-e artwork released.
• AI Music Generation: New features in stems for greater control over music production.
• College Football Video Game Development: AI used in rapid development of college football game assets using player likeness.

Conclusion

• The week saw significant advancements in AI models across the industry.
• Anticipation for utilizing and testing these new tools upon returning from vacation.
• Invitation to subscribe for further AI news and updates.