Anthropic can already retrain Mythos with "a fraction of the compute originally needed," says former researcher Coxon
Former Anthropic researcher Jacob Coxon claims that technical progress at the company is accelerating, citing internal data suggesting Mythos can be retrained using significantly less computing power than previously required. Coxon, who left the organization after reviewing these efficiency gains, has publicly argued for a pause in the development of frontier models. This disclosure highlights an ongoing industry debate regarding the pace of AI advancement and the potential risks associated with rapidly improving model training capabilities.
Covered by 1 source
- TThe Decoder↗Matthias BastianNov 14