newsfilter
← Latest News
General Catalyst·

The Data That Makes Frontier Models Smarter

Summary

  • Proximal, co-founded by Calvin Chen, Justus, & Navid, is a research partner helping frontier AI labs & coding startups build high-quality, complex training data.
  • The company emerged from a realization that existing data sources were too simple for advanced reinforcement learning, which requires diverse & difficult tasks to drive model growth.
  • Calvin Chen's entrepreneurial background includes dropshipping enablement in high school that generated millions, competitive golf, & a pivot from a failed Y Combinator batch to Proximal.
  • Proximal distinguishes itself by treating data creation as research & engineering rather than standard labeling, involving credible experts to design tasks & verify outputs.
  • The firm's RL-based training approach focuses on creating simulated environments where models solve complex problems like end-to-end Stripe integration to improve capabilities.
  • Proximal plans to expand its services beyond frontier labs to help enterprises with niche use cases that fall out of distribution for general models.
  • The company rapidly built a technical team of 14 to 22 engineers in India by applying San Francisco-level rigor & respect to attract top talent.
  • Calvin Chen emphasizes that model progress is bottlenecked on data quality & complexity, not just compute power or GPU availability.
  • The company aims to scale its capacity by recruiting 50 to 100 of the world's best researchers & engineers to accelerate model development.