"The Model Proposes, the Scholar Disposes: Building Trustworthy AI Instruments for Biblical and Hebrew Studies"

Large language models can already read Hebrew, Greek, Aramaic, Syriac, and Latin, but a model that answers questions is not yet an instrument that serves scholarship. This presentation reports on five interoperating open tools, BibCrit, SHEBANQ MCP, Peshitta Constellations, the Aramaic Root Atlas, and the Polyglot Concordance, that place the LLM at deliberately different points in the philological workflow: as data filter, hypothesis generator, query translator, grounded interpreter, and draft author. Across all five, three architectural invariants preserve scholarly authorship: deterministic verification surrounds every model output; machine-enforced provenance distinguishes corpus fact from model claim; and limitations, including measured cross-run instability, are published rather than hidden. Live demonstrations over the Masoretic Text, Septuagint, Qumran scrolls, Peshitta, and Targumic corpora will show what can responsibly be offloaded to models today and what cannot. The presentation closes with an argument for strong, durable partnerships between scholars and builders: each of these instruments became trustworthy only where expert correction flowed back into it. In an age of capable frontier models the field's scarcest resource, expert judgment, becomes more valuable, not less.

Jossi Fresco Benaim is an independent researcher and software builder working at the intersection of the digital humanities and Semitic philology. After a successful career as a technology executive and entrepreneur, he now explores the application of AI across different areas of human knowledge. Over the past year he has built and released five open research tools for biblical and Hebrew studies, BibCrit, SHEBANQ MCP, Peshitta Constellations, the Aramaic Root Atlas, and the Polyglot Concordance, all freely accessible in the browser, published with DOIs on Zenodo, and documented with open APIs. His work focuses on how large language models can be embedded in rigorously verifiable architectures that keep the scholar, not the model, as the author of record.

large language models; biblical textual criticism; Hebrew Bible; computational philology; Aramaic and Syriac corpora; digital humanities; scholarly provenance; open science