How much is 56 times 37? Language models often make mistakes in these types
of difficult calculations. This is usually explained by their inability to
perform complex reasoning. Since language models rely on large training sets
and great memorization capability, naturally they are not equipped to run
complex calculations. However, one can argue that humans also cannot perform
this calculation immediately and require a considerable amount of time to
construct the solution. In order to enhance the generalization capability of
language models, and as a parallel to human behavior, we propose to use special
'thinking tokens' which allow the model to perform much more calculations
whenever a complex problem is encountered.

当语言模型遇到复杂的计算问题时，由于其不能执行复杂的推理，它们通常会出现错误。为了增强语言模型的泛化能力，并与人类行为相对应，我们提出使用特殊的 “思考标记”，使模型在遇到复杂问题时能够执行更多计算。

语言建模的思维标记

Thinking Tokens for Language Modeling

Most knowledge graph use cases are data-centric, focusing on representing
data entities and their semantic relationships. There are no published success
stories to represent large-scale complicated business logic with knowledge
graph technologies. In this paper, we will share our innovative and practical
approach to representing complicated U.S. and Canadian income tax compliance
logic (calculations and rules) via a large-scale knowledge graph. We will cover
how the Tax Knowledge Graph is constructed and automated, how it is used to
calculate tax refunds, reasoned to find missing info, and navigated to explain
the calculated results. The Tax Knowledge Graph has helped transform Intuit's
flagship TurboTax product into a smart and personalized experience,
accelerating and automating the tax preparation process while instilling
confidence for millions of customers.

本文介绍了如何使用知识图谱技术代表大规模繁琐的美国和加拿大收入税合规逻辑，包括计算和规则。使用这种方法可以帮助 TurboTax 产品自动化和加速税务准备过程，同时为数百万客户带来智能化和个性化的体验，实现税务产品的转型。