AI News: A Massive Week For AI Advancement!

Introduction

I'm wrapping up my vacation here in Colorado, almost ready to head home, but I haven't let that stop me from keeping up with the AI news. This week has been monumental for AI, possibly the biggest week we've had this summer. Let's dive into all the exciting updates.

Meta Releases Llama 3.1

The most significant news is Meta's release of Llama 3.1, an upgraded model of Llama 3. It now comes in three sizes: 8 billion, 70 billion, and 405 billion parameters. Larger models usually offer better reasoning and math capabilities. Llama 3.1 also excels in areas like tool usage, multilingual capabilities, complex reasoning, and coding.

Meta's benchmarks show Llama 3.1 outperforming other state-of-the-art models like GPT-4 Omni and Claude 3.5 Sonet in several areas. The open-source nature of Llama 3.1 means developers can customize and fine-tune it to their specific needs. However, if you have more than 700 million monthly active users, you need to request a license from Meta.

Accessibility and Platforms

Running the 405 billion parameter model locally is beyond most people's computational capacity, but you can test these models on various platforms like AI, WhatsApp, Instagram Messenger, and Facebook Messenger. Grock and Perplexity also support Llama 3.1 models for those looking to experiment.

Mistral AI and More

Mistral AI released Mistral Large 2, a 123 billion parameter model, which outperforms the Llama 3.1 70 billion parameter model in math performance. Mistral Large 2 edges out other models in code generation for several programming languages. Apple also released 7 billion and 1.4 billion parameter models, claiming they outperform the Mistral 7B and are close to Llama 3. Google's Gemini model got an upgrade to Gemini 1.5 Flash, offering improvements in reasoning and image understanding, and expanding the token limit for free users to 32,000.

OpenAI Announcements

OpenAI has introduced the ability to fine-tune GPT-4 for free for up to 2 million training tokens a day until September 23rd. They also announced Search GPT, a new prototype for a better search experience. This provides a similar functionality to Perplexity, with a waitlist available for early testers.

AI Video Innovations

Luma AI released "Loop" in Dream Machine, allowing for infinite animation loops. Another advancement in AI video came from Cing AI, which is known for producing impressive text-to-video results. Runway faced controversy over allegedly training on YouTube videos without permission, raising questions about ethics and legality in training data sourcing.

Stability AI's Stable Video 4D

Stability AI released Stable Video 4D, a model that transforms a single object video into multiple novel view videos from different angles, available on Hugging Face.

Adobe and Leonardo

Adobe introduced new AI features in Illustrator, such as filling shapes with complex patterns. Leonardo rolled out a team collaboration feature, making it easier for multiple creators to work together.

AI in Game Development

EA Sports used AI to scan 11,000 college athletes for the reintroduction of NCAA Football. The use of AI accelerated the process, allowing for highly detailed and realistic player representations.

Keywords

• Meta
• Llama 3.1
• Mistral AI
• Google Gemini
• OpenAI
• Search GPT
• AI Video
• Cing AI
• Stability AI
• Adobe AI
• Leonardo
• AI in gaming

FAQ

1. What are the new models released by Meta?

   • Meta released Llama 3.1 in three sizes: 8 billion, 70 billion, and 405 billion parameters.

2. What is the significance of Mistral AI's new model?

   • Mistral AI released a 123 billion parameter model, Mistral Large 2, which outperforms Llama 3.1 in several benchmarks.

3. How can I access and test Llama 3.1 models?

   • Platforms like AI, WhatsApp, Instagram Messenger, Facebook Messenger, Grock, and Perplexity support Llama 3.1 models for testing.

4. What new features did Google Gemini introduce?

   • Google Gemini upgraded to Gemini 1.5 Flash, enhancing reasoning and image understanding and increasing the token limit to 32,000 for free users.

5. What is Search GPT by OpenAI?

   • Search GPT is a new prototype aiming to improve search functionality, allowing users to get answers with sources similar to Perplexity.

6. What advancements were made in AI video generation?

   • Luma AI released Loop for infinite animation, and Cing AI further improved text-to-video generation. Stability AI released Stable Video 4D for novel view video generation.

7. How did Adobe enhance its AI capabilities?

   • Adobe introduced features in Illustrator to fill shapes with complex patterns and enhance image generation capabilities.

8. What is significant about the use of AI in EA Sports' game development?

   • EA Sports used AI to scan over 11,000 college athletes' faces for the NCAA Football game, improving realism and detail.

Feel free to explore all these exciting developments as AI continues to make significant strides across various fields!