DPO

The researchers present ComPO, a new alignment paradigm that uses comparison oracles to extract directional information from preference pairs. This method effectively mitigates the risks of likelihood displacement and model verbosity seen in traditional direct alignment methods.

Researchers have developed SecOPD, a fine-tuning technique that uses token-level feedback to protect AI agents from prompt injection attacks. This method significantly outperforms previous state-of-the-art defenses by precisely identifying and penalizing malicious tokens during training.

Researchers have developed SecOPD, a fine-tuning technique that uses token-level feedback to protect AI agents from prompt injection attacks. This method significantly outperforms previous state-of-the-art defenses by precisely identifying and penalizing malicious tokens during training.

This paper proposes APPO, a method to enhance the training of autonomous LLM agents through more granular credit assignment. It moves beyond trajectory-level rewards by identifying and branching at specific high-impact decision points within the reasoning process.