Parrot Mind: Towards Explaining the Complex Task Reasoning of Pretrained Large Language Models with Template-Content Structure
This work addresses the open question of LLM reasoning mechanisms for researchers in AI and machine learning, offering a theoretical explanation that could inform model design and interpretation.
The authors tackled the problem of explaining how pretrained large language models (LLMs) learn complex reasoning tasks by proposing a template-content structure that reduces the generation space from exponential to linear levels, and they validated this structure experimentally in current LLMs.
The pre-trained large language models (LLMs) have shown their extraordinary capacity to solve reasoning tasks, even on tasks that require a complex process involving multiple sub-steps. However, given the vast possible generation space of all the tasks, how the pretrained model learns the reasoning ability remains an open question. We firstly propose that an intrinsic structural constraint on the generated sequence of language-based reasoning -- we called it template-content structure (T-C structure) -- is the key to explain why LLMs can solve a large number of complex reasoning problems with limited training data by showing this structure can reduce the possible space from exponential level to linear level. Furthermore, by generalizing this structure to the hierarchical case, we demonstrate that models can achieve task composition, further reducing the space needed to learn from linear to logarithmic, thereby effectively learning on complex reasoning involving multiple steps. We provide both examples and formal theory of our T-C structure. We also experimentally validate the existence of the T-C structure in some current LLMs and its effectiveness for reasoning.