Andrew Ho, a former OpenAI researcher, has warned that frontier AI labs may face a 50% valuation drop. He has since started a company to sell high-end reinforcement learning datasets, with $700k in locked OpenAI equity.
TL;DR
- Andrew Ho, a former OpenAI researcher, warns of a potential 50% drop in frontier AI lab valuations.
- Ho has started a company to sell high-end reinforcement learning datasets, focusing on long-horizon scientific reasoning and statistical analysis.
- Ho's locked OpenAI equity, worth $700k, cannot be sold until after the IPO and lockup period.
What happened
Andrew Ho left OpenAI after eight months and announced on X that he was starting a company to sell high-end reinforcement learning datasets to frontier AI labs.
Ho posted advice for his former colleagues, recommending they take liquidity if eligible for tender offers, as he believes the valuation of frontier labs is overvalued and could drop by 50%.
Ho has approximately $700k in OpenAI shares that he cannot sell until after the IPO and lockup period, leaving him 'stuck' with the equity.
Why it matters
Ho's warnings highlight concerns about the sustainability of current AI lab valuations and the potential for a significant drop in value.
His new company focuses on creating datasets for tasks that models still struggle with, such as long-horizon scientific reasoning and statistical analysis, addressing a critical gap in AI development.
The situation puts Ho at odds with some Silicon Valley figures who believe in the potential of recursive self-improvement (RSI) to drive exponential growth in AI capabilities and revenue.
Key facts
- Andrew Ho left OpenAI after eight months to start a company selling high-end reinforcement learning datasets.
- Ho has $700k in locked OpenAI equity that cannot be sold until after the IPO and lockup period.
- Ho warns of a potential 50% drop in the valuation of frontier AI labs.
- His new company will focus on long-horizon scientific reasoning and statistical analysis datasets.
- Ho believes that competition from cheaper models is forcing frontier labs into a Red Queen's race, where they must spend more to stay competitive.
- He is skeptical about the potential of recursive self-improvement (RSI) to drive exponential growth in AI capabilities and revenue.
- Ho suggests that the winners in the current AI landscape are Nvidia and Micron, who sell chips to all the labs in the race.
Context
The AI industry is currently experiencing a significant selloff, with the Nasdaq 100 entering correction territory. This has raised concerns about the valuations of frontier AI labs and their ability to generate revenue.
The debate over the potential of recursive self-improvement (RSI) is a hot topic in the AI community, with some believing it could lead to exponential growth in AI capabilities and revenue, while others, like Ho, are skeptical.
The situation highlights the challenges faced by AI labs in balancing the need for continuous innovation and the financial pressures of staying competitive in a rapidly evolving market.