Portable Semantics, Private Dialects: Reuse and Negative Transfer in Latent Communication Between Language-Model Cells
Narcis Marincat (independent researcher) tests whether independently trained societies of language-model cells that communicate through latent packets share one packet language, and finds that they do not, and that an inherited communication interface can hurt later learning.
Ask this paper
Interoperability audit: Across all 30 ordered pairs of six independently trained societies, with a preregistered raw, orthogonal, linear and nonlinear alignment ladder, only one same-initialization pair is fully interoperable. All 26 cross-initialization directions fail at every alignment level.
Failure location: A source-span control places strict zero-shot failure in interpreting and executing the new operator instructions.
Negative transfer: Reinitializing only the packet reader, writer and output head of the globally trained interface raises final depth-three accuracy from 0.169 to 0.857.
Reuse: Inherited interfaces never beat fresh-interface controls by the preregistered 0.10 margin. Conclusions are limited to a 17-state near-transfer setting.
Abstract
In shared-genome language-model societies, restricted evidence visibility favors reusable, value-indexed latent packet interfaces, whereas the sole high-performing globally visible model in the parent study learned an episode-entangled code. This companion study asks whether independently trained societies share one packet language, where strict zero-shot transfer fails, and whether inherited interface state helps or harms later learning. First, a leakage-controlled causal interoperability audit over all 30 ordered pairs of six independently trained restricted societies -- under sealed held-out structure and a preregistered raw/orthogonal/linear/nonlinear alignment ladder -- shows the six semantically similar interfaces do not form one raw language: one same-initialization pair is exactly interoperable in both directions, a second shows asymmetric partial compatibility, and all 26 cross-initialization directions fail every frozen alignment rung. Second, within the tested decomposition and a single sealed source formulation, a source-span control localizes strict zero-shot failure to interpretation and execution of the new operator instructions. Third, in a matched adaptation factorial, the globally trained communication interface acts as a severe negative-transfer prior: reinitializing only the packet reader, writer, and mouth raises final depth-three accuracy from 0.169 to 0.857. Fourth, across two restricted checkpoints and two independently frozen target streams each, inherited interfaces never exceeded fresh-interface controls by the preregistered 0.10 margin. All primary conclusions are bounded to a near-transfer 17-state setting; the negative-transfer factorial concerns one globally visible parent-cohort checkpoint, while an appendix adds a post hoc tagged-global twin case study.