Here is a study of failure modes of using LLMs to control robots. I thought it was interesting and might spur some thoughtful discussion. The focus is on controlling robots, but I think a lot of these failure modes apply to agents in general. The article, published in "International Journal of Social Robotics", contains a fairly detailed evaluation of model biases.
Here is a study of failure modes of using LLMs to control robots. I thought it was interesting and might spur some thoughtful discussion. The focus is on controlling robots, but I think a lot of these failure modes apply to agents in general. The article, published in "International Journal of Social Robotics", contains a fairly detailed evaluation of model biases.