Your Rules. Your AI.

Artificial intelligence grows more advanced every month and now reaches into nearly every industry. Yet complex and highly regulated organizations still can't trust it with their most important work. Model developers set their models' values, guardrails remain vague and loosely enforced, and most responsible AI programs never make it past the policy document. The alignment risks are simply too high.

Ordinary Wisdom closes that gap. Our platform and advisory service let you control any AI model according to your company's standards, regulatory requirements, and values. You can get started immediately, without rebuilding your existing systems or hiring AI safety specialists. You define the rules, and we make every model follow them.

Key Features

The patent-pending Ordinary Wisdom platform controls and evaluates behavior of any LLM (proprietary or open source) using multiple machine learning and data science techniques. Our platform includes:

• Granular control of LLM output through
  • Portfolios of documents (e.g. policies, regulatory) that inform AI output
  • Constitutions of rulesets and values that that control AI output
  • Advanced ML and post-generation control methods
• Evaluation of outputs against specified values, including comparison and scoring methods, and LLM as a judge
• Application programming interface allowing controlled LLM output to be used in any application.

We are steadily improving and extending our platform, including adding methods for post-generation control of model outputs, and to optimize control methods.

See our documentation for a detailed description of current features.

Use Cases and Implementation

Ordinary Wisdom is effective across a wide range of sectors, including:

• Regulated industries, e.g., financial services, healthcare, legal
• Agent deployment at scale, e.g., robotics, defense
• Brand owners, e.g. to establish a distinctive brand voice

We engage actively with customers on implementation of specific use cases. Please contact us to discuss.