As a Member of Technical Staff (Research) at Quadrillion Labs, you'll help train our next-generation model for research creativity. This is genuinely hard: very few people have post-trained models at the scale of several hundred billion parameters, and we're doing it with a lean team and a set of ingenious ideas - accumulating research data and training on it in an unusually data-efficient way. If you're excited about turning every arXiv paper on the internet into an RL environment, or building fast RL loops between user data and model training, this role may be for you.
What you'll do- Build infrastructure for model training. You'll build our ultra-efficient infrastructure for scraping and accumulating data, turning it into RL environments, and post-training our frontier model - especially on extremely long-horizon tasks.
- Develop RL techniques. No one has figured out how to train a model that's truly good at research yet. You can help us change that with mysterious RL techniques.
- Construct and run evaluations and benchmarks. Build evaluations for complex, long-horizon tasks and measure both frontier models and our own against them - so we know what's working and where to push next.
Required qualifications- Experience training or benchmarking large, modern models
- Ability to move quickly in a zero-to-one environment
Preferred qualifications- Experience training hundreds-of-billions-of-parameters frontier models
About this roleWe work in-person 5 days a week out of our office in Midtown Manhattan. We cover generous benefits for employees, including medical, dental, and vision insurance, lunch and dinner covered, and other varied stipends.
We pride ourselves on a company culture that values a diversity of thoughts and opinions.
