We imagine a world where anyone can create their own model by teaching it, just like you would a person. Our AI adapts to your preferences and history and improves the more you use it. The only limits are your own creativity and taste.
We don’t see the world through the lens of pre-training or post-training. Our models are always training. Continuous retraining requires developing infrastructure for retraining models on an online basis—at the scale of hours, minutes, or even seconds.
Long-term, we’re building new ways for people to provide feedback to AI that compounds over time. Day to day, we develop novel algorithms that enable models to update themselves from new data. Doing this well requires rethinking every layer of the stack, from data to algorithms to architecture.
Good ideas don’t originate in a vacuum. Even our most abstract research is grounded in the desire to eventually improve our product experience. Solving continual learning will require innovation beyond theoretical formulations or simulated benchmarks.
We’re a small, focused team of researchers and engineers working at the frontier of learning and memory. We include some of the world experts in parameter-efficient architectures, language model memory, and continual learning.
We’re looking for exceptional people who want to do their best work. Get in touch at hello@engram-lab.com.