Google's Deep Mind announced Gemini 1.5, enhancing AI language models with a mixture of experts architecture.

• Gemini 1.5 is an upgrade from last week's Gemini Ultra, featuring a mixture of experts architecture for efficient processing.
• The model can handle up to 1 million tokens, which translates to about 750,000 words of input and output text.
• This capability allows for uploading extensive texts, such as the entire Harry Potter series, and asking detailed questions about it.
• Gemini 1.5 has improved multi-modality understanding, demonstrated by analyzing a 44-minute silent Buster Keaton movie.

OpenAI unveiled Sora, an AI that can generate realistic videos, and shared more details in a research paper.

• Sora can create videos up to 60 minutes long with lifelike quality, challenging the status quo of AI capabilities.
• The research paper includes technical explanations, demo videos, and details on how Sora works.
• Sora's abilities include generating videos from image prompts, extending videos, merging videos seamlessly, and more.
• OpenAI added a tool for interacting with video variations of Sora's capabilities on their research paper page.

The week was packed with various other AI news and updates, including from Meta, Nvidia, and Stability AI.

• Stability AI released Stable Cascade, a model that excels at generating art with legible text.
• Nvidia's Chat with RTX, an offline large language model interface, allows querying documents and YouTube videos.
• Meta's V JEA is a video model that predicts interactions in videos, aiming to advance machine intelligence.
• 11 Labs introduced a feature to monetize personal voices, allowing others to use them in exchange for rewards.