InFerActive: Interactive Tree-Based Exploration of LLM Sampling for Safety Evaluation
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866913175366008832 |
|---|---|
| author | Hwangbo, Junhyeong Lee, Soohyun Jeon, Hyeon Jang, Kyochul Cheong, Minsoo Yu, Youngjae Seo, Jinwook |
| author_facet | Hwangbo, Junhyeong Lee, Soohyun Jeon, Hyeon Jang, Kyochul Cheong, Minsoo Yu, Youngjae Seo, Jinwook |
| contents | Even LLMs that appear safe during evaluation can still produce harmful responses in deployment. Because stochastic sampling yields different responses to the same prompt, low-probability harmful outputs can still reach users at scale. Common human evaluation workflows generate many random samples per prompt and review them in static spreadsheets. The practice scales poorly, forcing evaluators to repeatedly reread near-duplicate prefixes. To address this, we present InFerActive, an interactive system that visualizes sampling results as a navigable tree of readable phrases, allowing evaluators to filter, explore, and extend the generation space on demand. InFerActive utilizes breadth-first sampling, a novel tree construction procedure that matches the harmful-response coverage of random sampling while requiring up to 5.0x fewer samples. Two controlled user studies (N = 12 each) demonstrate that InFerActive significantly improves evaluation efficiency and coverage over both spreadsheet and basic tree baselines. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_10234 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | InFerActive: Interactive Tree-Based Exploration of LLM Sampling for Safety Evaluation Hwangbo, Junhyeong Lee, Soohyun Jeon, Hyeon Jang, Kyochul Cheong, Minsoo Yu, Youngjae Seo, Jinwook Human-Computer Interaction Artificial Intelligence Even LLMs that appear safe during evaluation can still produce harmful responses in deployment. Because stochastic sampling yields different responses to the same prompt, low-probability harmful outputs can still reach users at scale. Common human evaluation workflows generate many random samples per prompt and review them in static spreadsheets. The practice scales poorly, forcing evaluators to repeatedly reread near-duplicate prefixes. To address this, we present InFerActive, an interactive system that visualizes sampling results as a navigable tree of readable phrases, allowing evaluators to filter, explore, and extend the generation space on demand. InFerActive utilizes breadth-first sampling, a novel tree construction procedure that matches the harmful-response coverage of random sampling while requiring up to 5.0x fewer samples. Two controlled user studies (N = 12 each) demonstrate that InFerActive significantly improves evaluation efficiency and coverage over both spreadsheet and basic tree baselines. |
| title | InFerActive: Interactive Tree-Based Exploration of LLM Sampling for Safety Evaluation |
| topic | Human-Computer Interaction Artificial Intelligence |
| url | https://arxiv.org/abs/2512.10234 |