Jacob Coxon, a former Anthropic pretraining researcher, says a slide shown to investors during the IPO roadshow played a major role in his resignation. It highlighted superexponential gains in pretraining compute efficiency, a commercial argument for Anthropic but also a warning sign for him. Coxon argues that such advances could lead to self-improving AI systems arriving sooner than expected, with potentially existential consequences. His short tenure came with a significant financial sacrifice, as he forfeited unvested equity before an IPO seeking to raise approximately $64 billion.
Source: Read the original article


