Skip to main content
BusinessModelsResearch

Anthropic can already retrain Mythos with "a fraction of the compute originally needed," says former researcher Coxon

Source: The Decoder (opens in a new tab) · Matthias Bastian

Intel Summary

Former OpenAI and Anthropic researcher Jacob Coxon stated that Anthropic can retrain its Mythos model using a fraction of the compute originally required, citing a slide from the company's IPO roadshow as a reason for his departure. Coxon asserted that Anthropic's algorithmic efficiency is improving at an accelerating pace and reiterated calls to pause frontier model development.

Why It Matters

Rapidly diminishing compute requirements for retraining advanced models point to steep algorithmic efficiency gains across leading AI labs. If accurate, lower compute barriers accelerate frontier iteration cycles while intensifying debates around safety pauses, market concentration, and the commoditization of human-level AI systems.

Organizations & Entities