analiza aiqbench
This research introduces HEIMAT, a heuristic-based framework designed to automatically debias large language models without the prohibitive computational costs or manual annotation requirements of existing methods. By leveraging simple prompt templates and fine-tuning techniques that minimize prediction divergence, the approach successfully mitigates cultural biases while preserving natural language understanding capabilities. The work addresses critical challenges in AI safety by offering a scalable solution for reducing social harm generated during model interactions.
A Heuristic Perspective on Debiasing Language Models
arXiv:2608.00622v1 Announce Type: new Abstract: Language models (LMs) often acquire various biases during pre-training and may express them in interactions, potentially causing social harm. Existing methods often rely on counterfactual augmentation or representation projection. These strategies remain limited in practice due to their high computational costs and difficulty in scaling to larger models. Additionally, many of these strategies require manual data annotation, narrowing their scope to specific cultures and bias categories. To overcome these limitations, we propose HEIMAT, a HEurIstic-style autoMATic debiasing framework for LMs. HEIMAT consists of two main steps: bias disclosure and debiasing fine-tuning. In the first step, it uses simple templates to construct heuristic prompts, which are applied to reveal model biases and generate corresponding context prompts. In the second step, it fine-tunes the model by minimizing the Jensen-Shannon divergence of predictions on these context prompts to reduce bias. Extensive experiments show that HEIMAT effectively mitigates bias in different cultures while maintaining the model's natural language understanding (NLU) performance.
Abstract:Language models (LMs) often acquire various biases during pre-training and may express them in interactions, potentially causing social harm. Existing methods often rely on counterfactual augmentation or representation projection. These strategies remain limited in practice due to their high computational costs and difficulty in scaling to larger models. Additionally, many of these strategies require manual data annotation, narrowing their scope to specific cultures and bias categories. To overcome these limitations, we propose HEIMAT, a HEurIstic-style autoMATic debiasing framework for LMs. HEIMAT consists of two main steps: bias disclosure and debiasing fine-tuning. In the first step, it uses simple templates to construct heuristic prompts, which are applied to reveal model biases and generate corresponding context prompts. In the second step, it fine-tunes the model by minimizing the Jensen-Shannon divergence of predictions on these context prompts to reduce bias. Extensive experiments show that HEIMAT effectively mitigates bias in different cultures while maintaining the model's natural language understanding (NLU) performance.
Computation and Language (cs.CL)
arXiv:2608.00622 [cs.CL]
(or arXiv:2608.00622v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2608.00622 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 252 cuvinte
Subiecte conexe
Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
AI Alignment Forum · 3 aug. 2026
What can language models teach us about understanding?
LessWrong · 9 oct. 2026
Reviewing failure modes in science as we prepare for AI researchers
LessWrong · 8 oct. 2026
Learning What to Forget: Improving LLM Unlearning via Learned Token-Level Importance
arXiv · 6 iun. 2026