Fine-tuning LLMs on structured tasks inspired by maladaptive behaviors produces stable, context-general shifts in next-token distributions and response tendencies consistent with altered behavioral priors.
Advances in Neural Information Processing Systems , year=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Modeling Pathology-Like Behavioral Patterns in Language Models Through Behavioral Fine-Tuning
Fine-tuning LLMs on structured tasks inspired by maladaptive behaviors produces stable, context-general shifts in next-token distributions and response tendencies consistent with altered behavioral priors.