Essential AI
To overcome humanity's enormous challenges, we need powerful AIs that can mimic and amplify our greatest strength – the ability to solve unseen problems. We see a line of sight to a bright future where humans and machines solve problems that are out of reach for either of them alone.
We believe the future of AI should not be dictated by a select few deciding which problems are worth solving. Innovation thrives when knowledge is accessible, collaboration is encouraged, and everyone has the tools to contribute.
At Essential AI, we're committed to shaping a future where progress is driven by open science - where knowledge is shared, not siloed.
Optimizers
Our research team has demonstrated that Muon, a lightweight second-order optimizer, achieves better compute-time tradeoffs than AdamW for large language model training. When combined with muP (maximal update parameterization), it provides a practical recipe for efficient large-scale pretraining.
Infrastructure
Our infra team has developed a novel layer sharding strategy for scaling the Muon optimizer to large language model training. This approach dramatically reduces computational overhead while maintaining optimization benefits, making Muon practical for training models at scales like Llama 405B.
Careers
Join our team of talented individuals working on cutting-edge AI technologies. We are based in SF and work onsite 5 days a week. If you don't see an open role but still greatly want to contribute towards building next-generation AI research and development, please reach out to hiring@essential.ai
We are based in SF and are building a world-class multi-disciplinary team of engineers, researchers, designers, and sales and product experts who are excited to solve hard real-world AI problems. We are excited to partner with March Capital and Thrive Capital, with participation from AMD, Franklin Venture Partners, Google, KB Investment, and NVIDIA.