LATIDIA · Investigación

Actualización de creencias conscientes de la procedencia en los bucles de retroalimentación de la IA humana

arXiv:2610.13008v1 Tipo de anuncio: nuevo Resumen: En la interacción humana-AI repetida, una sola observación puede citarse, resumirse y devolverse a un modelo como múltiples informes, haciendo que la repetición parezca corro independiente

La noticia

arXiv:2610.13008v1 Announce Type: new Abstract: In repeated human-AI interaction, a single observation can be quoted, summarized, and returned to a model as multiple reports, making repetition look like independent corroboration. We study whether language models can use supplied report lineage to update according to evidence provenance rather than report count. We introduce Latent Quotient Evidence Training (LQET), a relational objective that combines sparse Bayesian supervision with constraints that encourage stable predictions when reports add no new evidence and consistent updates to the same new evidence across relay histories. On a controlled binary benchmark, LQET reduces new-evidence update RMSE by 41-44% and log-odds drift after eight relay steps by 89-91% relative