A deep dive into the foundational shifts Karpathy highlights for 2025: reinforcement learning from verifiable rewards (RLVR) driving massive cheap optimization, the rise of ‘thinking time’ traces and jagged, task-optimized intelligence, and the birth of vibe coding—guiding powerful AI with plain language. We explore the new LLM app layer that turns lab models into deployed professionals, on-device agents like Claude Code, and the multimodal LLM GUI that blends text, visuals, and code. Finally, we discuss the economic dynamics, evolving benchmarks, and the future jobs vibe coding may unlock in the year ahead.

Note: This podcast was AI-generated, and sometimes AI can make mistakes. Please double-check any critical information.

Sponsored by Embersilk LLC