I'll caveat this article by beginning with a caveat. Artificial Intelligence for widespread commercial use, especially for specialized applications such as law, finance, medicine, and tax, is very much in its infancy. There are major leaps in the underlying technology on a constant basis. Because of this, the theoretical application of the technology is not always congruent with the practical outcomes.

In its simplest form, artificial intelligence (AI) built on large language models (LLMs) are trained on massive amounts of data which can be interacted with by, in our case, human users. Interacting with these technologies allows users to ask questions, perform analysis, automate tasks, and in general augment our own knowledge, skills and abilities.

An important benefit of this technology is the potential to apply memory as contextual data points to a single thread. A thread begins with a single question, prompt, ask, or command to the LLM. In our case, using TaxGPT Research, we might ask "My client is a software developer, do they qualify for the R&D credit?". This question establishes the thread, along with some preliminary context, such as the client's industry and the applicability of the R&D credit.

Once the thread is established, this context remains present in the background as you continue to ask additional questions. As you ask more questions, more context is generated and saved. Potentially, with enough questions and data points, the AI builds and retains an exhaustive contextual library as it applies to each thread.

Using TaxGPT's client profiles, found as Individual and Business Client options on TaxGPT Research's landing page, give the AI a running start with regards to context. Profiles allow users to include common, important tax-related data points such as client type, jurisdiction(s), filing status, income/revenue, deductions, industry, etc. Once profiles are complete, a thread is created bearing the client's name.

This client thread now contains preliminary context from which the AI uses as you interact with it. This contextual library continues to expand as you ask questions related to the thread, gathering more data points and further building out the client profile for you, becoming increasingly more valuable as more questions are asked.

To revisit the preface, this does not always work perfectly in practice. The AI can get locked into the thread context to the point you can't ask hypotheticals or ask it to depart from important data points. In some cases, TaxGPT Research's general chat function may be better suited for your queries.