5 Questions
What type of cognitive abilities are highly desirable for machine learning systems in mathematical problem solving?
Learning new skills and inferring abstract rules from limited data
What type of models may be able to support advanced cognitive abilities if their training data samples these abilities widely enough?
Large language models with billions of parameters
What game is used in the text to address the limitation of machine learning models to learn truly novel information and exhibit out of distribution generalization?
Sudoku
What strategy did human participants learn from a narrow range of training examples that showed strong out of distribution generalization in solving Sudoku?
Hidden Single strategy
What was required for small-scale transformers to generalize the Hidden Single strategy out of distribution?
Continuously interleaving the narrowly sampled Hidden Single examples with examples of component strategies spanning the entire distribution of puzzle instances
Explore the intersection of large language models and mathematical problem solving by examining their ability to learn new skills and infer abstract rules from limited data. Understand the potential of large language models in supporting advanced cognitive abilities and learning truly novel information.
Make Your Own Quizzes and Flashcards
Convert your notes into interactive study material.
Get started for free