+To clarify, there are several global initiatives to address and reduce biases in LLMs, particularly those related to data, like the demographic’s ones. These biases can often be minimized or managed using LLMs themselves, employing strategies such as guardrails or reinforcement learning from human feedback methods. Adhering to the **HPE AI Ethics Principles**, we are also committed to mitigate these biases. However, we recognize that there is a distinct category of biases: the cognitive biases, which, unlike data biases, receive less focus in academic research and have fewer mitigation strategies available. Cognitive biases, such as confirmation bias, have profound effects on human interaction, potentially leading to polarization also within small social units like teams or departments. In response to this challenge, we are also dedicating efforts towards understanding and addressing cognitive biases. We are fostering environments that **encourage diverse perspectives by creating 'shared memory' spaces** within projects. In these spaces, employees can contribute with their individual chat memories, facilitating the collection of varied viewpoints and promoting openness and dialogue.
0 commit comments