Let's Regulate This AI Stuff?

Demis Hassabis, CEO and co-founder of DeepMind, proposes the creation of a US-based frontier AI standards body modeled after FINRA. This organization would oversee the review of new AI models up to 30 days before their release, with the authority to evaluate and potentially slow down the deployment of risky models. Hassabis emphasizes the need for global regulation as AI systems approach the potential of artificial general intelligence and the singularity. He has been actively promoting this initiative with support from US and European officials, aiming to establish a robust regulatory framework before year's end.

IBM reported a slight increase in Q2 revenue but fell short of expectations, with shares dropping over 20%. The decline is attributed to a shift in customer spending from IBM's products to chips and servers, driven by shortages in semiconductor supply caused by the global AI-driven build-out of data centers. This demand surge has increased costs across the industry and slowed IBM's software and mainframe sales. CEO Arvind Krishna noted that IBM was unprepared for the rapid customer shift and supply chain issues, impacting its transition towards a high-growth software company.

Spotify has introduced 'Talk to Spotify,' a new feature in beta for Premium users aged 18+ in the US, Ireland, and Sweden. The feature allows users to control their music, explore listening history, and learn about songs and podcasts via voice or typed commands within the app. This integration of AI aims to enhance personalized music experience, making it easier for users to navigate and discover content through natural language interactions.

Kalshi has launched a new market-based tool that uses prediction markets to track the future costs of AI resources like GPUs, storage, and memory. This forward curve provides a market-based expectation of future compute expenses, aiding companies in hedging and risk management as demand for AI infrastructure continues to grow. As compute becomes a critical commodity, this tool offers valuable insights into pricing trends and future investment needs.

The podcast discusses the emerging concept of 'world models,' which are AI systems capable of simulating environmental interactions to predict future events and understand physical and perceptual consistency in real time. Unlike traditional language models, world models aim for continuous, synchronous interaction, and have significant implications for robotics, virtual environments, and physical AI. Several companies have raised substantial funding around this concept, viewing world models as a potential next phase in AI development that extends beyond the capabilities of large language models.