Nick Saraev and Jack Roberts discuss reports of books being scanned for AI training and ask how access to scarce human-created materialTraining data is the collection of examples and signals used to adjust an AI model's parameters so it learns useful patterns. affects competition. They distinguish the destruction of a purchased copy from the much stronger, unverified claim that all surviving copies are being removed, while raising concerns about cultural preservation.
Nick Saraev and Jack Roberts discuss reported activity around Ilya Sutskever's Safe Superintelligence and speculate about alternatives to current model training. Their discussion of architecture and online learningContinual learning lets an AI system acquire new knowledge or skills over time while retaining what it learned earlier. is conjecture, not evidence of a demonstrated breakthrough. A separate discussion considers how chip accessAI compute infrastructure is the hardware, facilities, networks, power, cooling, storage, and software used to train and run AI models. and export restrictions shape competition.
Nick Saraev and Jack Roberts caution against treating model usage rankingsEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria. as a direct measure of intelligence or business success. Price, distribution and workload can affect token volumeToken volume is the total number of input, output, cached, or reasoning tokens an AI workload processes over a defined period or task.; they argue that useful results and customer outcomes are more meaningful measures.
Watch on YouTube




