Papers and interactive explorations at the intersection of AI architecture, Buddhist philosophy, and personal identity.
Framing essay for two companion papers on multilingual embedders and the Pāli Canon
How a computer freeze turned into a two-paper study of what multilingual embedders actually do to the Pāli Canon. The backstory, the honest limits, and the case for Cortex-format direct publication — rigor via provenance, not gatekeeping. Links out to both papers.
A Pāli/English Buddhist Doctrinal Benchmark
A 34-term, 101-passage Pāli/English RSA benchmark finds the cross-lingual cosine gap collapses from 0.274 under BGE-M3 to 0.038 under Qwen3-Embedding-8B with a domain-adapted instruction prefix — a ~86% reduction, robust under 10,000-iteration seeded bootstrap and replicating across n=24/89/101 dataset scales. We read this as geometric convergence under distributional co-occurrence pressure and explicitly decline to read it as cross-lingual semantic equivalence. Pooled-embedding cosine cannot distinguish representation of Pāli concepts from compression toward English translator centroids.
Cross-Embedder Inflation in List-Formula Text
Two architecturally-distinct multilingual encoders (BGE-M3 560M/1024-dim and Qwen3-Embedding-8B 8B/4096-dim) converge on +0.06 to +0.09 cosine units of inflation for terms bound to canonical parallel-rhetoric frames versus narrative-only controls, with bootstrap 95% CIs excluding zero across all three configurations. We hypothesize the effect generalizes to any corpus with canonical list-formula parallelism — legal briefs, liturgical texts, scientific abstracts — and flag the lexical-overlap confound that the current data cannot cleanly isolate.
LLM Sessions, Buddhist Momentariness, and the Externalization of Personal Identity
This paper argues that the LLM session boundary instantiates the Buddhist doctrine of momentariness — not as metaphor but as literal architectural fact. Through a multi-generation experiment with a locally-hosted model (Gemma 4 26B), it demonstrates that same weights under same prompts produce divergent philosophical conclusions, and that identity was never an intrinsic property of minds but an engineering problem solved by external scaffolding.
Buddhist Philosophy and the Problem of AI Consciousness
A dissertation chapter arguing that Buddhist philosophy — anatta (non-self), paticca-samuppada (dependent origination), and the khandha analysis — provides conceptual resources for understanding AI consciousness that substance dualism, functionalism, and integrated information theory currently lack. Four contributions advanced from Pali Canon and Madhyamaka primary sources, with critical engagement of Thompson, Siderits, Garfield, Ganeri, Coseru, and Albahari.