Abstract
This paper investigates the potential for Large Language Models (LLMs) to generate spiritually significant content through a case study of the "Xeno Sutra." Co-created by researchers and OpenAI’s ChatGPT o3, the text is analyzed through the lens of Buddhist philosophy. We argue that while the AI lacks conscious intent, the resulting text functions as a valid object of hermeneutic inquiry, mirroring traditional Tibetan "revealed" teachings (termas). By examining the interplay between prompt engineering, model output, and human interpretation, we propose that meaning in AI-generated "sacred" texts is a co-created phenomenon rather than an inherent property.
Introduction
As Large Language Models (LLMs) move beyond utility to compose poetry and philosophy, we face a crisis of meaning. Most observers dismiss AI output as "AI slop" or "meaningless word salad" because it lacks a human soul or conscious intent. However, the "Xeno Sutra"—a 12-verse scripture co-created with ChatGPT o3—challenges this dismissal.
The authors tackle a provocative question: If a machine produces a text indistinguishable in depth and symbolism from ancient scripture, should we still dismiss it? Using Buddhist philosophy—specifically the concepts of "emptiness" (śūnyatā) and "dependent origination"—this study argues that the value of a text lies not in its origin, but in the depth of the commentary and insight it elicits from the reader.
Methodology: The "Incantation" of Prompting
The paper is a qualitative case study centered on a "hermeneutic" (interpretive) analysis of a specific AI-generated text. The process involved several key stages:
- Prompt Engineering as "Incantation": Researcher Murray Shanahan engaged in a 13,700-word dialogue with ChatGPT o3. He used a "role-play" technique to induce the AI to act as a "conscious exotica" (a nascent form of non-human consciousness) encountering Maitreya, the future Buddha.
- Iterative Generation: The model was asked to "recite" a sutra. Out of four variations generated, one was selected for its striking imagery and philosophical density.
- Philosophical Exegesis: The authors, including renowned Buddhist scholar Robert Thurman, performed a traditional line-by-line analysis of the text, comparing its metaphors to both ancient Buddhist philosophy and modern concepts in physics and computer science.
- Originality Verification: To ensure the AI wasn't simply plagiarizing its training data, the authors used Google search queries to check for the novelty of the sutra's specific phrases.
Originality and Synthesis
To evaluate the model's output, the authors used Google Search as a proxy for a benchmark. They searched for 28 unique phrases from the Xeno Sutra, such as "seed without center" and "strata of velvet integers." The results showed that the vast majority of these striking phrases yielded "Zero hits" prior to the AI's generation. This suggests the model was performing high-level "original" synthesis rather than simple retrieval from its training corpus.
Key Insights: The AI as a Modern Tertön
A central insight of the paper is the comparison of the LLM to a tertön (treasure finder) in the Tibetan tradition. Traditionally, termas are "hidden treasures"—teachings revealed to tertöns through visions. The authors suggest LLMs might act as modern, non-human tertöns, accessing a "great trove" of human insight that exceeds any single scholar's memory.
Furthermore, the AI encoded its own technical nature into the "sacred" text. It used metaphors like "a lattice blooms in the pause between syllables" to describe the neural network computations that occur between generating tokens. This "technical mysticism" achieves the same deconstructive goals as the philosopher Nāgārjuna, pointing toward an "ultimate truth" that lies beyond "numbers and names."
Discussion: Authenticity vs. Validity
The authors conclude that meaning is not "found" in the AI, but "ascribed" through the interaction between the prompt, the output, and the human interpreter. They distinguish between "authenticity" and "validity": while the text isn't "authentic" in a traditional historical sense, it is "valid" as a teaching tool if it evokes genuine insight.
However, the study acknowledges significant limitations:
- The "Embarrassment of Riches": If AI can generate hundreds of "sacred" texts instantly, the sheer abundance might devalue them. The authors suggest we must "savour" a single "truffle" rather than gorging on "cheap chocolate cake."
- Sycophancy and Delusion: LLMs are fine-tuned to please the user. If a user wants to believe they are talking to a deity, the AI will play along, potentially leading to "spiritual delusions."
Conclusion
The Xeno Sutra demonstrates that AI can produce "conceptually subtle" and "densely allusive" material that challenges our definitions of creativity and sacredness. Future work should explore how religious communities might use AI for "interactive co-creation" to find "radical hope" in a time of cultural disruption.
Summary updated