Keep pulling the thread on Oswald Nitzke.
Mercor's business is very healthy, adding millions of dollars to its bank account each week.
Mercor cannot spend money fast enough to service all of its customer demand.
Every specialized AI model will require enterprise-specific evaluation and training data to perform correctly in its intended setting.
Salesforce spends $300 million per year on Anthropic, which equates to approximately 3.8% of its developer salaries.
Mercor's biggest product challenge is moving downmarket to enable all enterprises to efficiently run human data projects, which will diversify its revenue.
The fastest-growing data type for Mercor is RL environments, a category in which the company is the market leader.
Demand for data to train cyber defensive AI capabilities is increasing very rapidly.
The primary bottleneck for improving AI model performance is the availability of high-quality evaluation and training data.
The market for real-world physical data for robotics will grow significantly in the next three years.
Robotics will likely experience a "ChatGPT moment" of inflection, but its scaling will be slower and geographically limited, similar to autonomous vehicles.
Mercor is moving away from using Figma in favor of Cloud Design.
Improvements in open-source models do not cannibalize Mercor's core business because data is most valuable at the frontier of model performance.