Dualformer: Controllable Fast and Slow Thinking by Learning with Randomized Reasoning Traces
View PDF
HTML (experimental)
Abstract:In human cognition theory, human thinking is governed by two systems: the fast and intuitive System 1 and the slower but more deliberative System 2. Recent studies have shown that incorporating System 2 process into Transformers including large language models (LLMs), significantly enhances their reasoning capabilities. Nevertheless, models that purely resemble System 2 thinking require substantially higher computational costs and are much slower to respond....
Read more at arxiv.org