PERSON DIRECTORY
Mike Knoop
Mike Knoop appears in 2 indexed conversations across Gradient Dissent, Machine Learning Street Talk. This directory brings every appearance, source, TL;DR, digest, and transcript into one searchable feed.
ARC Prize Version 2 Launch Video! [Francois Chollet, Mike Knoop]
ARC-AGI-2 resets the AGI benchmark around tasks humans solve within two attempts but frontier systems largely cannot, with pretrained models near 0%, o3 around 4–5%, and average human performance around 60%. o3’s apparent test-time search marks a possible architectural discontinuity, yet its thousands-of-dollars-per-task compute and ARC-AGI-2 gap keep efficiency unresolved as ARC Prize 2025 tests whether open-source teams can close it.
R1, OpenAI’s o3, and the ARC-AGI Benchmark: Insights from Mike Knoop on the Gradient Dissent Podcast
ARC-AGI v1 marks a sharp capability break: GPT-4-class systems scored roughly 4%, versus o3 at 75% and an expensive high-compute configuration at 85%.Inference-time reasoning and search appear to drive the shift, but whether pure RL can reach o3-level adaptation and unlock reliable enterprise agents remains unresolved.