Marco Mascorro
Showing 1–3 of 3 transcripts.
- a16z59 min
The 2045 Superintelligence Timeline: Epoch AI’s Data-Driven Forecast
Yafah Edelman, David Owen, Marco Mascorro, Erik Torenberg, Kirsten, Jeff
Speakers analyze current AI spending as a signal of immediate value rather than a bubble, noting that while rapid compute growth and data center scaling are underway, future profitability hinges on whether large model investments yield proportional returns. The discussion forecasts rapid economic integration with potential GDP gains of 30% under AGI scenarios, alongside a 20-30% risk of a sudden unemployment spike that could trigger immediate government intervention. Looking toward technical milestones, the event projects significant breakthroughs in mathematical problem-solving and robotics within five years, while warning that traditional benchmarks will soon saturate as the industry shifts toward measuring real-world system performance.
- a16z42 min
Google DeepMind Lead Researchers on Genie 3 & the Future of World-Building
Jack Parker-Holder, Shlomi Fruchter, Anjney Midha, Marco Mascorro, Justine Moore, Erik Torenberg
Google DeepMind has released Genie 3, a research preview that generates interactive, photorealistic 3D worlds in real-time from text prompts to support navigation and control. Built by integrating insights from three internal projects, the model introduces spatial memory for one-minute object persistence and emergent physical reasoning to distinguish it from previous video generation systems. While currently limited to visual simulation without audio, Genie 3 aims to bridge the sim-to-real gap for robotics and agent training by providing diverse, high-fidelity environments free from physical data collection risks.
- a16z27 min
DeepSeek, Reasoning Models, and the Future of LLMs
Guido Appenzeller, Marco Mascorro
DeepSeek R1 is an open-weight reasoning model from China that achieves top-tier performance by combining Multi-Head Latent Attention, Group Relative Policy Optimization, and a 256-expert MoE architecture to generate complex thought chains. The development team overcame early behavioral failures through a low-cost, self-supervised pipeline utilizing 800,000 verifiable traces and rule-based verification to produce responses up to 10,000 tokens long for roughly $5.5 million in base training costs. This breakthrough has shifted industry focus toward test-time compute and local deployment, enabling state-of-the-art reasoning on consumer hardware while bypassing traditional bottlenecks associated with human-labeled data.