How it works

Authored knowledge - recorded lectures, papers, textbooks - belonging to a single professor is manually screened and organized for conversion into training data. (On its own, AI will impose generic patterns, creating slop.) After initial guidance, the AI will recreate a professor's intellectual logic while grounding content in the source materials instead of the anonymous data guiding today's large language models.

Caffeinated Professor lets professors control their AI, instead of having AI invent the professors' thought patterns.

Models

  • Lecturer: Recorded audio/video and slides ingested into the model, typically from a single class. Subsequent AI interactions governed narrowly by the lecture substance and rhetoric.
  • Lecturer + textbook: Recorded audio/video and slides from a single class ingested into the model, along with the course textbook (regardless of author). AI interactions governed narrowly by the lecture substance and rhetoric as enriched by the textbook. (Lecture takes priority over text.)
  • Textbook: Text ingested into the model. AI interactions governed narrowly by book content.
  • Textbook + author: On top of the textbook model, the author may incorporate their own supplementals, including audio and video of course lectures, lecturer's written work on the subject, and similar.
  • Textbook + lecturer: On top of the textbook model, the professor who adopts the textbook may incorporate their own supplementals to bend the textbook into their unique presence. Supplementals may include audio and video of course lectures, lecturer's written work on the subject, and similar. The effect is a textbook AI agent personalized to the specific human lecturer. ("Identity" of AI professor becomes ambiguous but constrained between textbook's priority over lecturer.)