Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructions
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2023
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866909140889108480 |
|---|---|
| author | Bianchi, Federico Suzgun, Mirac Attanasio, Giuseppe Röttger, Paul Jurafsky, Dan Hashimoto, Tatsunori Zou, James |
| author_facet | Bianchi, Federico Suzgun, Mirac Attanasio, Giuseppe Röttger, Paul Jurafsky, Dan Hashimoto, Tatsunori Zou, James |
| contents | Training large language models to follow instructions makes them perform better on a wide range of tasks and generally become more helpful. However, a perfectly helpful model will follow even the most malicious instructions and readily generate harmful content. In this paper, we raise concerns over the safety of models that only emphasize helpfulness, not harmlessness, in their instruction-tuning. We show that several popular instruction-tuned models are highly unsafe. Moreover, we show that adding just 3% safety examples (a few hundred demonstrations) when fine-tuning a model like LLaMA can substantially improve its safety. Our safety-tuning does not make models significantly less capable or helpful as measured by standard benchmarks. However, we do find exaggerated safety behaviours, where too much safety-tuning makes models refuse perfectly safe prompts if they superficially resemble unsafe ones. As a whole, our results illustrate trade-offs in training LLMs to be helpful and training them to be safe. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2309_07875 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructions Bianchi, Federico Suzgun, Mirac Attanasio, Giuseppe Röttger, Paul Jurafsky, Dan Hashimoto, Tatsunori Zou, James Computation and Language Training large language models to follow instructions makes them perform better on a wide range of tasks and generally become more helpful. However, a perfectly helpful model will follow even the most malicious instructions and readily generate harmful content. In this paper, we raise concerns over the safety of models that only emphasize helpfulness, not harmlessness, in their instruction-tuning. We show that several popular instruction-tuned models are highly unsafe. Moreover, we show that adding just 3% safety examples (a few hundred demonstrations) when fine-tuning a model like LLaMA can substantially improve its safety. Our safety-tuning does not make models significantly less capable or helpful as measured by standard benchmarks. However, we do find exaggerated safety behaviours, where too much safety-tuning makes models refuse perfectly safe prompts if they superficially resemble unsafe ones. As a whole, our results illustrate trade-offs in training LLMs to be helpful and training them to be safe. |
| title | Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructions |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2309.07875 |