Holy crap! Anthropic's AI self-evolution speed has scared internal researchers into resigning! Jacob Coxon, a former pre-training researcher at Anthropic, publicly shared an internal chart purportedly from Anthropic's IPO roadshow. This is one of the key reasons he decided to leave. Over the past two years, the computing power required to train an AI model of equivalent capability has been decreasing dramatically, and the rate of decrease is accelerating. According to him, retraining Mythos now requires only a fraction of the computing power it did during initial training. Next step: AI will soon begin to research how to improve its own training efficiency. The next step may be AI improving algorithms to make the next generation of AI smarter; the next generation of AI will then improve algorithms again, further accelerating the next generation. AI trains AI, AI improves AI, AI accelerates the creation of even stronger AI. This is recursive self-improvement.