Hippocratic AI develops safety-focused large language models with a focus on non-diagnostic, patient-facing applications. The majority of language models undergo pre-training using the common crawl of the Internet, which might contain inaccurate and deceptive information.