CASA: Concept-Axis Self-Adjustment. A Closed Loop for Continuous Self-Correction of Language Models

This paper turns a raw research intuition into a concrete algorithm. The intuition: while running,a language model should be able to notice that its representation of an entity disagrees withverifiable knowledge—its own freshly generated statement, an implication of an ontology, or anexternal reference—and apply a small displacement of that representation along the relevantconcept axis; accumulated displacements should then be periodically consolidated into weights.The result would be a closed loop of self-improvement that never retrains on a corpus. Wedevelop the idea into CASA (Concept-Axis Self-Adjustment): a formal setup, update ruleswith a trust region, a two-level memory (edit cache → weights), stability conditions, an honestlist of failure modes, and a falsifiable MVP plan for a single GPU.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-19
DOI
https://doi.org/10.5281/zenodo.24362541
Primary Topic
Topic Modeling
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

CASA: Concept-Axis Self-Adjustment. A Closed Loop for Continuous Self-Correction of Language Models

Nedilko Aleksandr Yuryevich
Zenodo (CERN European Organization for Nuclear Research)
Topic Modeling
preprint

CASA: Concept-Axis Self-Adjustment. A Closed Loop for Continuous Self-Correction of Language Models

Nedilko Aleksandr Yuryevich
preprint en

Abstract

This paper turns a raw research intuition into a concrete algorithm. The intuition: while running,a language model should be able to notice that its representation of an entity disagrees withverifiable knowledge—its own freshly generated statement, an implication of an ontology, or anexternal reference—and apply a small displacement of that representation along the relevantconcept axis; accumulated displacements should then be periodically consolidated into weights.The result would be a closed loop of self-improvement that never retrains on a corpus. Wedevelop the idea into CASA (Concept-Axis Self-Adjustment): a formal setup, update ruleswith a trust region, a two-level memory (edit cache → weights), stability conditions, an honestlist of failure modes, and a falsifiable MVP plan for a single GPU.

Zenodo (CERN European Organization for Nuclear Research)
Topic Modeling
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon