Legendary VC Steve Jurvetson looks ahead at neutral networks, Tesla, nuclear power, and more | E2193

What It Covers

Legendary venture capitalist Steve Jurvetson discusses Moore's Law's 130-year trajectory, the shift from CPUs to GPUs for AI workloads, nuclear energy misconceptions, analog computing's potential, and why AI alignment resembles parenting more than programming.

Key Questions Answered

• •Moore's Law Reality: Computing price-performance has doubled annually for 130 years across mechanical, relay, vacuum tube, transistor, and integrated circuit eras, delivering a thousand billion billion-fold improvement. This exponential continues despite economic disruptions, with algorithmic improvements now doubling yearly alongside hardware advances.
• •GPU Dominance Over Intel: Intel lost Moore's Law leadership 15 years ago by focusing on backward-compatible single processors rather than fine-grained parallel architectures. Nvidia's GPUs and custom ASICs from Google, Amazon, and OpenAI now drive AI workloads through massive parallel computation, fundamentally better suited for matrix operations than traditional CPUs.
• •Analog Computing Breakthrough: Companies like Mythic store eight bits of information in a single transistor versus traditional eight-transistor-per-bit designs, enabling 1000x better power efficiency. This biomimetic approach mirrors brain function with massively parallel, low-power computation, potentially enabling trillions of edge AI devices in consumer products at costs below plastic buttons.
• •Nuclear Energy Misinformation: Germany's nuclear shutdown costs $220 million daily to Russia and causes 1,100 excess deaths annually from fossil fuel pollution. Fukushima caused zero radiation deaths, while coal kills 4 million people yearly from particulates. The conflation of nuclear weapons with nuclear power, amplified by 1979 anti-nuclear concerts, created irrational policy decisions.
• •AI Alignment as Parenting: AI systems cannot be reverse-engineered, controlled, or proven safe like traditional engineering products because they are inherently inscrutable complex systems. Regulation should focus on policing inputs and outputs rather than internal alignment, similar to parenting teenagers versus programming deterministic code. Truth-seeking algorithms beat mind control approaches.