SafeTune: Search-based Harmfulness Minimisation for Large Language Models
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910201251102720 |
|---|---|
| author | d'Aloisio, Giordano Williams, David Annunziata, Giusy Fei, Zhiwei Di Marco, Antinisca Sarro, Federica |
| author_facet | d'Aloisio, Giordano Williams, David Annunziata, Giusy Fei, Zhiwei Di Marco, Antinisca Sarro, Federica |
| contents | The widespread adoption of Large Language Models (LLMs) raises concerns about the potential harmfulness of their responses. In this paper, we first investigate the harmfulness of responses from four general-purpose LLMs. Next, we propose SafeTune, a multi-objective search-based approach to mitigate harmfulness while increasing response relevance through hyperparameter tuning and system prompt engineering. Our initial evaluation shows that SafeTune significantly reduces the rate of harmful responses generated by Qwen3.5 0.8B and increases prompt-response relevance (both with a large effect size). Among the parameters we explore, we also find that encouraging greater repetition in responses is most impactful in reducing harmfulness while increasing relevance. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_07709 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | SafeTune: Search-based Harmfulness Minimisation for Large Language Models d'Aloisio, Giordano Williams, David Annunziata, Giusy Fei, Zhiwei Di Marco, Antinisca Sarro, Federica Software Engineering The widespread adoption of Large Language Models (LLMs) raises concerns about the potential harmfulness of their responses. In this paper, we first investigate the harmfulness of responses from four general-purpose LLMs. Next, we propose SafeTune, a multi-objective search-based approach to mitigate harmfulness while increasing response relevance through hyperparameter tuning and system prompt engineering. Our initial evaluation shows that SafeTune significantly reduces the rate of harmful responses generated by Qwen3.5 0.8B and increases prompt-response relevance (both with a large effect size). Among the parameters we explore, we also find that encouraging greater repetition in responses is most impactful in reducing harmfulness while increasing relevance. |
| title | SafeTune: Search-based Harmfulness Minimisation for Large Language Models |
| topic | Software Engineering |
| url | https://arxiv.org/abs/2605.07709 |