Build better AI
We believe that AI agents can transform the world for better or for worse. AI agents and models are neither inherently good nor inherently evil nor value-neutral; they embody the values of their developers, the biases in the data on which they were trained, and the preferences of their users. For AI to transform the world for the better, people need agents that they can trust. Our mission simply is to help AI developers build trustworthy agents.
Culture
We are inspired by the values we inherited from Amazon, Intel, the original Hewlett-Packard, and AT&T Labs, and this is reflected in our culture.
Focus on the Customer:
We start with known customers and work backwards to build a viable offering. Because we value our user’s attention, we build high-quality products with well-designed interfaces. We earn and keep customer trust with every interaction.
Take Ownership:
We create long-term value for our customers, our employees, and our investors. We work to get more done with less because constraints inspire invention.
Learn Relentlessly:
We move fast, fail fast, and learn from our mistakes. We have an inexhaustible appetite for knowledge, competence, and performance. Being on the bleeding edge of a new field, we are always on the lookout for new tools and techniques, and use them quickly to improve our products.
Trust and Respect:
We approach everyone with the belief that they want to do a good job and will do so given the context and tools. We attract highly capable, diverse, and innovative people. We expect them to achieve ambitious goals. We help everyone grow to their full potential at the company. We make time for coaching, collaboration, and candid communication. We have no tolerance for jerks, no matter how brilliant.
Dive Deep:
As practitioners, we keep asking Why and know that every answer is only provisionally correct. As leaders, we stay connected to the details and audit ourselves and others objectively. We benchmark ourselves against the best in the field.
Innovate:
We must be creative because the metrics and mechanisms of trustworthy agents simply do not exist. We look for new ideas from everywhere, and are not limited by the notion it was “not invented here”. We have a seemingly infinite opportunity to make AI easier to use for everyone everywhere.
Deliver Results:
We get stuff done. We believe research is "done" not when a paper is published but when the idea is built into a testable prototype for its first customer. We believe code is done not when it’s complete but when it’s tested, documented, and used by customers in production. We believe a sale is done not when the deal is closed but when the customer renews their contract.
Careers
You will build upon the latest research methods in trustworthy ML and generative AI to build and deliver cloud services for AI developers who are customizing agents and LLMs. You will collaborate with applied scientists to build capabilities for testing, tuning, and deploying agents and LLMs with security and safety. Your day-to-day responsibilities include:
- Solve problems facing enterprises that want to deploy agents and LLMs in production
- Implement ML techniques to detect LLM vulnerabilities to attacks
- Implement ML techniques to assess LLM propensities for harm
You will be responsible for research and development of novel machine learning techniques to improve the security and safety of large language models. You will continuously survey SOTA research in adversarial ML, red-teaming, data curation, training, and fine-tuning LLMs; you will pursue independent research to develop patents and write papers for peer-reviewed publication at top-tier conferences; you will develop small-scale prototypes (production-quality code) and collaborate with ML engineers and cloud services developers to transform your prototypes into large-scale cloud services. Importantly, you will focus obsessively on the needs of customers -- AI developers at enterprises and startups who are customizing LLMs to build agents -- and let them guide the direction of your research.