• Starter AI
  • Posts
  • AI's societal dangers, DeepMind's MC-ViT, LLMs progress in spatial sound

AI's societal dangers, DeepMind's MC-ViT, LLMs progress in spatial sound

Sam Altman has issued a warning.

Author

Sthefania
February 13, 2024

Hello, Starters!

When it comes to AI, there's a lot to learn: prompts, methods, papers, and plenty of words that may seem weird at first glance. I understand; I've been there! That's why we're here to help!

Here’s what you’ll find today:

  • Sam Altman warns of AI's potential dangers

  • Google DeepMind presents MC-ViT

  • LLMs are learning to differentiate spatial sounds

  • Microsoft is catching up with Amazon

  • Nvidia unveils its Ada Generation GPU

  • And more.

⛔️ Sam Altman warns of AI's potential dangers (3 min)

During a video conference at the World Government Summit in Dubai, OpenAI's CEO, Sam Altman, confessed that "AI risks" keeping him up at night do not revolve around "killer robots," but rather concern "very subtle societal misalignments" that could lead to worst-case scenarios.

Altman consistently emphasises OpenAI's commitment to the development of responsible AI technologies. While acknowledging that the industry is still in the "early stages of discussions," he believes that at a certain point, a concrete global action plan for AI regulations should be put in place.

A research study from Google DeepMind and Cornell University has introduced MC-ViT, a Memory-Consolidated Vision Transformer that enhances the capabilities of transformer-based video encoders for a better understanding of extended video sequences.

This approach diverges from traditional methods where challenges due to lengthy material tend to occur. Instead, MC-ViT focuses on key information from earlier segments, similar to how human memory consolidation works. This way, the compressed “memory” improves processing efficiency while providing accurate results with fewer resources.

Large language models are not as good as humans at identifying a certain type of sound and differentiating where it comes from. That's why a team of researchers has presented BAT, the first spatial, audio-based LLM designed to decipher sounds in a 3-D environment.

BAT represents a leap toward truly multimodal AI systems, thanks to its impressive precision in determining types of sound, direction, and distance. It takes LLM development to a whole new level by enabling them to not just understand words as we do, but also similarly process sound.

📊 Microsoft's AI growth is surpassing Amazon's, as reports suggest that the Azure cloud infrastructure is gradually narrowing the gap with leading AWS. This surprising twist is happening due to Microsoft's collaboration with OpenAI and its continuous focus on AI development.

💻 Nvidia aims to boost AI adoption with the introduction of the RTX 2000 Ada Generation GPU. This graphics processing unit was designed to enhance performance, versatility, and AI capabilities. It serves as the ideal hardware solution for businesses and professionals seeking to optimise their workflows.

What did you think of today's newsletter?

Login or Subscribe to participate in polls.

Thank you for reading!