Ariel Elboim
Independent researcher, Israel
I study what learning systems actually compute, by taking celebrated measures and results apart until they reduce to simpler, testable mechanisms: in language models, in models of how languages emerge, and in brain-imaging measures of integration. Where possible, tests are fixed before the data are seen, and code and data are released with each paper.
Alongside research I run an AI implementation practice for businesses, after more than ten years of building technology products.
Papers
-
Rethinking the Synergistic Core in Resting-State fMRI: Persistence and Connectivity Reconstruct the Synergy–Redundancy Ranking in Prospective Tests
-
What Do Zero-Data Self-Play Models Add Beyond an Off-the-Shelf Compressor? A Byte-by-Byte Test
-
Why This Language? Historical Symmetry Breaking through Cultural Transmission
-
Risk Normalization Does Not Imply Scale Invariance: Costs, Harmonic Width, and Eligibility in Futures Trading Rules