Evaluating AI Reasoning Paths
An AI model is being trained to produce step-by-step solutions. The training process rewards solutions based on the sum of the log-probabilities of each step being correct (a higher, i.e., less negative, sum is better). Both of the following solution paths for the same problem arrive at the correct final answer. Based only on the provided reward mechanism, which path will be more strongly reinforced during training, and what does this preference reveal about the training objective?
0
1
Tags
Ch.5 Inference - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Evaluation in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
An AI model is being trained to generate step-by-step reasoning. The training process provides a reward for each complete reasoning path, calculated by summing the log-probabilities of each individual step being deemed 'correct'. A higher (i.e., less negative) total reward is better. Consider the following four reasoning paths generated by the model, along with the log-probability of correctness for each step. Which path will be most strongly reinforced during the training process?
Evaluating AI Reasoning Paths
Critique of Log-Probability Reward Signals