算法效率
Methodological gains that reach equal model capability with less compute, shaping training cost curves.
Former Anthropic researcher Jacob Coxon says a chart shown during Anthropic's IPO roadshow reveals super-exponential gains in algorithmic efficiency, meaning Mythos could now be retrained with a fraction of its original compute. He argues for a one-year pause on frontier model development to let competitors catch up.
Former Anthropic researcher Jacob Coxon says the company can already retrain its Mythos model with "a fraction of the compute originally needed," and cites a steep compute-growth chart from Anthropic's IPO roadshow as a key reason he left. The claim hints that algorithmic efficiency may sharply lower frontier training costs, though it rests on a personal account without official figures.