SafeTune uses multi-objective search to tune LLM hyperparameters and prompts, significantly cutting harmful responses and raising relevance on Qwen3.5 0.8B with large effect sizes.
In: GECCO (2007)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SE 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
SafeTune: Search-based Harmfulness Minimisation for Large Language Models
SafeTune uses multi-objective search to tune LLM hyperparameters and prompts, significantly cutting harmful responses and raising relevance on Qwen3.5 0.8B with large effect sizes.