What happened
Matthew Harvey Sanders' company, Longbeard, developed Magisterium AI, a Catholic AI model, by scanning nearly 30,000 sacred works, including papal encyclicals and doctrinal writings, into a large language model dataset using robotic scanners and AI-powered software called Vulgate. Concurrently, Gloo evaluated leading AI models (Grok, OpenAI, Meta, DeepSeek) on human flourishing principles, finding scores over 80/100 for fact-based categories like finance but an average of 48/100 for faith, where models struggled with concepts such as sin and grace.
Why it matters
The emergence of highly specialised AI models like Magisterium AI demonstrates a clear mechanism for achieving depth and accuracy in complex, nuanced domains. For architects and founders, this indicates that general-purpose models face inherent constraints in handling abstract, non-factual domains, necessitating custom training on extensive, curated datasets for applications requiring specific worldviews or theological perspectives. This follows recent discussions around AI discernment, highlighting the need for domain-specific knowledge to move beyond generic outputs.
