Reducing Non-Normative Text Generation from Language Models
This addresses the issue of harmful or inappropriate text generation from AI models for users and developers, representing an incremental improvement in safety fine-tuning.
The paper tackled the problem of language models generating non-normative text by fine-tuning GPT-2 with a policy gradient reinforcement learning technique and a normative classifier, resulting in a reduction of non-normative text by 27-61% across datasets.
Large-scale, transformer-based language models such as GPT-2 are pretrained on diverse corpora scraped from the internet. Consequently, they are prone to generating non-normative text (i.e. in violation of social norms). We introduce a technique for fine-tuning GPT-2, using a policy gradient reinforcement learning technique and a normative text classifier to produce reward and punishment values. We evaluate our technique on five data sets using automated and human participant experiments. The normative text classifier is 81-90% accurate when compared to gold-standard human judgments of normative and non-normative generated text. Our normative fine-tuning technique is able to reduce non-normative text by 27-61%, depending on the data set.