OpenAI前员工刚跑路就喊话：要套现就赶紧套，别等IPO！

TL;DR - Former OpenAI employee Andrew Ho argues that employees should sell shares through tender offers rather than await an IPO, because frontier AI lab valuations assume highly uncertain long-term growth while compute and training costs keep rising.

• Public markets may discount AI labs that lack strong cash flow, profits, and near-term growth.
• Sustaining model leadership requires continual spending on GPUs, data, clusters, and training amid intense competition and falling API prices.
• Ho doubts current capability gains will quickly generalize from verifiable tasks like coding and math to complex real-world work.
• He expects high-quality reinforcement-learning data to become increasingly scarce and valuable as scaling returns slow.