Pierce Freeman and Richard Diehl Martinez discuss Recursive's Ripple 65B preview, contrasting its dense architecture with mixture-of-experts models and questioning whether its reported results justify a frontier label. They describe the company's reported training cost as a claim and note that they have not independently stress-tested the model.
The hosts discuss local opposition to data centers and the relationship between permitting, electricity, waterAI compute infrastructure is the hardware, facilities, networks, power, cooling, storage, and software used to train and run AI models. and broader AI regulation. Their election and market interpretations are presented as commentary.
They revisit experiments in which language models inspect or modify smaller neural networks' weightsWeight space learning treats the trained numerical parameters of neural networks as data a system can analyze, predict or transform.. They distinguish varied experimental setups and sandbox access from claims of autonomous recursive self-improvementRecursive self-improvement is the proposed process in which an AI system helps improve its own capabilities, then uses those improvements to support further advances., while discussing potential uses and limitations.
Watch on YouTube
