Episode Details

Move over, Chinchilla Scaling—the era of "Inference Scaling" has arrived. We rip the pipes out of the LLM-as-a-Verifier (LaV) framework to explore how "System 2" thinking and process-based supervision are allowing smaller models to outsmart the giants. From the promise of high-value latency to the dangers of "verification collapse" and AI steganography, we dive deep into the shift from bigger models to smarter thinking.