Google

Misalignment in Large Language Models stems from several key issues. One primary source is a model's failure to follow user instructions, a capability not inherently acquired during pre-training. Another significant cause is the model's tendency to generate harmful content or replicate societal biases that are present in its training data.

Sources of LLM Misalignment

Analyze the model's response in the following scenario. Identify two distinct sources of misalignment demonstrated in this interaction and explain how each source is represented in the model's output.

Analyzing Misalignment in a Customer Support LLM

Match each example of a large language model's output with the primary source of misalignment it demonstrates.

A user asks a large language model to write a short biography of a famous female scientist. The model's response disproportionately focuses on her personal life and relationships, while similar requests for male scientists result in biographies centered on their professional accomplishments. Which source of model misalignment does this scenario best illustrate?

Learn Before

Related