Papers

télos (τέλος): Preliminary Report 2 — Capability Divergence, Positional Encoding, and the Limits of Masked Diffusion Scaling↗

Following the initial scaling observations in Masked Diffusion Language Models (MDLMs) for code autocomplete, we complete our scaling sweeps across the 12.5M, 25M, and 50M parameter tiers. We identify a fundamental limitation in MDLM capacity allocation: past a critical,…