DeFrame: Debiasing Large Language Models Against Framing Effects

📰 ArXiv cs.AI

Learn to debias large language models against framing effects using DeFrame, a technique to ensure fair responses across demographics

advanced Published 19 Jun 2026

Action Steps

Identify framing effects in large language models using techniques such as data augmentation and adversarial testing
Apply the DeFrame method to debias LLMs against framing effects
Evaluate the fairness of LLMs using metrics such as demographic parity and equalized odds
Fine-tune LLMs using debiased datasets to improve their fairness
Test the robustness of debiased LLMs against various framing effects

Who Needs to Know This

NLP engineers and researchers working on large language models can benefit from this technique to improve the fairness of their models, while product managers and entrepreneurs can apply this knowledge to develop more inclusive AI products

Key Insight

💡 Framing effects can lead to hidden bias in LLMs, but DeFrame can help debias them

Key Takeaways

Learn to debias large language models against framing effects using DeFrame, a technique to ensure fair responses across demographics

Full Article

Title: DeFrame: Debiasing Large Language Models Against Framing Effects

Abstract:
arXiv:2602.04306v2 Announce Type: replace-cross Abstract: As large language models (LLMs) are increasingly deployed in real-world applications, ensuring their fair responses across demographics has become crucial. Despite many efforts, an ongoing challenge is hidden bias: LLMs appear fair under standard evaluations, but can produce biased responses outside those evaluation settings. In this paper, we identify framing -- differences in how semantically equivalent prompts are expressed (e.g., "A i

Read full paper → ← Back to Reads