Explaining the most common AI terms today.

Artificial intelligence (AI) terminology has now transcended the field of computer science and is gradually becoming part of everyday language. Many concepts that once only appeared in engineering textbooks are now frequently mentioned in work and daily life.

This article will explain the most common AI terms in a more easily understandable way, while also providing practical examples in accounting and business to help visualize how these technologies work.

What is Artificial Intelligence (AI)?

The concept of Artificial Intelligence was first coined in 1955 and became more widely known in 1956, although the idea of ​​machines capable of 'thinking' had existed much earlier.

According to the Merriam-Webster dictionary, AI is the ability of a computer system or algorithm to mimic intelligent human behavior.

Of course, this level of 'intelligence' has changed considerably over time with the development of technology. For example, if you look at modern automotive production lines today, the level of automation in the factory might have been considered 'extremely advanced AI' a few decades ago.

What is Generative AI?

Generative AI is a collection of AI technologies that have been continuously developing since 1956, but a major turning point occurred in 2017 when Google researchers published their famous paper titled 'Attention Is All You Need'.

This work introduces a mechanism that allows a computer to read an entire sentence or paragraph and then determine the importance of each word based on the words that appeared before it.

Simply put, AI will understand that the word 'bank' can have completely different meanings if the word 'river' appears before it instead of 'loan'. It all depends on the context.

This is actually quite similar to how accountants work on a daily basis. When auditors examine transactions for signs of fraud, they always place the data within the specific context of the business. A transaction might be normal at a building materials store but unusual at a religious organization.

What is the Large Language Model (LLM)?

Large Language Models — or LLMs — are the 'backbone' behind modern generative AI.

LLM contains a massive amount of training data collected from the internet. When a user enters a prompt, the generative AI applies probability and statistics to this data to generate a relevant response. This process is similar to an accountant taking data from a chart of accounts and creating a graph or analyzing trends to find outliers.

In other words, LLMs don't 'think' like humans, but predict the next token based on patterns learned from training data.

What is a prompt?

A prompt is simply the text that the user enters into the AI ​​tool.

Before the AI ​​can generate a response, the system must analyze the prompt to determine the context and the importance of each word. Only then does the AI ​​begin to generate a response based on that context.

Interestingly, when generating each subsequent word in the response, the AI ​​continues to use the context of the content it just created. This is similar to a client asking an accountant about their tax return. The accountant will consider the client's wording and combine it with the context from the tax return to provide an appropriate response.

What is a token?

Tokens are the basic units that AI can understand.

In many cases, a token is equivalent to a word, but sometimes it's just a part of a longer word. AI uses tokens in the order they appear to process training data and generate responses.

The context, reasoning, and prediction of AI are all built upon tokens.

What is temperature?

Temperature is a parameter that determines the degree of 'creativity' or 'randomness' of the AI ​​when choosing the next token.

Low temperatures will make AI responses more precise, rigid, and focused on accuracy. Meanwhile, high temperatures—for example, 0.9°F—will make the test model more likely to be less common, suitable for creative work but also more prone to errors.

For accounting or factual data processing tasks, lower temperatures are generally safer.

What is Foundation Model?

Foundation models are the frameworks that many AI companies use to build their products.

These are familiar names like OpenAI's GPT, Google's Gemini, and Anthropic's Claude.

Instead of building LLMs from scratch—which is extremely expensive—many businesses will license or integrate foundation models from these vendors to add AI features to their products.

What is alignment?

Alignment is the ability of AI to maintain alignment with the goals set by the user.

For example, if the prompt says 'help me manage my inbox but don't delete emails', the AI ​​alignment is to assist in processing emails but absolutely not to delete them.

Throughout the process, AI may receive various additional instructions, but the original core objective must remain unchanged.

What is fine-tuning?

Fine-tuning is the process of refining a foundation model to better suit a specific task or feedback style.

For example, a tax research platform might fine-tune its model to provide a consistent, structured response for tax memos or legal research.

In other words, fine-tuning not only helps AI answer more accurately, but also helps shape how those answers are presented.

What is Retrieval-Augmented Generation (RAG)?

RAG is a method of providing AI with more specific data so that the system can focus on that information source when answering questions.

A simple example is a business storing all its internal SOPs as text and then connecting them to AI. Employees can ask the AI ​​about company procedures, and the system will respond based on internal documents instead of general data on the internet.

More complex RAG systems may use charts of accounts or enterprise databases to help classify transactions or make recommendations.

What is a Context Window?

AI can only 'remember' a limited amount of text at any given time. This amount of text is called the context window.

The context window encompasses all prompts and responses that appear in a conversation with the AI. For example, some of Anthropic's models have a context window of around 200,000 tokens—equivalent to nearly two average novels.

When the context becomes too long, the model will begin compressing or summarizing older content to continue processing. However, this process sometimes causes the AI ​​to 'forget' important alignments or information that was mentioned earlier.

What is Model Context Protocol (MCP)?

Model Context Protocol (MCP) is a method that allows AI to connect with existing tools and systems. For example, using a mobile app to connect with a General Ledger accounting system. In this case, the phone acts as the foundation model, and the app acts as the MCP server. The two can communicate with each other because they use a common 'language' — in this case, the MCP protocol.

This is also why MCP is considered one of the most important platforms for modern AI agents and automation workflows.

What is Artificial AI?

Agentic AI is a type of AI that can perform actions on its own instead of just generating text responses.

Basic examples include sending emails, entering data, or working with business software.

AI agents can independently determine how to achieve user-defined goals by combining several of the technologies mentioned above. For example, a user provides a PDF invoice and asks the AI ​​to input the data into accounting software like Xero. The agent will automatically read the PDF, connect to Xero's MCP server, check if the vendor exists, and then input the transaction into the system.

This is the direction of development that is causing AI to gradually shift from a 'question-answering tool' to a 'system capable of autonomously managing workflows'.

---

Many AI terms that were once only found in engineering textbooks have now become common concepts in everyday work.

As AI continues to be more deeply integrated into businesses, understanding concepts like LLM, RAG, MCP, or agentic AI will become increasingly important—not just for developers, but also for accountants, managers, and office staff in general. Because in the next few years, AI may no longer be 'technology for tech people,' but will become a very normal part of how people work every day.