- 23
- February
Over the past 2-3 years, terms like AI, LLM, GPT, Prompt, and RAG have appeared everywhere — in business news, executive seminars, and software advertisements. Many of you may have tried ChatGPT or Google Gemini and thought "interesting, but I do not fully understand the big picture." This article explains everything from the fundamentals to how to get started in your organization — practical knowledge that executives can immediately put to use.
AI vs Machine Learning vs Deep Learning vs Generative AI
Many people confuse these terms. In reality, they are concentric circles from largest to smallest:
| Level | Term | Meaning | Example |
|---|---|---|---|
| Outermost Circle | AI (Artificial Intelligence) | Technology that enables computers to "think" or "make decisions" like humans | Product recommendations, self-driving cars, voice assistants |
| Second Circle | Machine Learning (ML) | AI that learns from data without manually programming every rule | Spam detection, sales forecasting, customer segmentation |
| Third Circle | Deep Learning (DL) | ML using multi-layered Neural Networks capable of learning complex patterns | Image classification, language translation, speech recognition |
| Innermost Circle | Generative AI | DL that can "create new things" — whether text, images, audio, or code | ChatGPT, Claude, Midjourney, GitHub Copilot |
In summary: Generative AI is a subset of Deep Learning, which is a subset of Machine Learning, which in turn is a subset of AI — they are not the same term, but rather "descendants" of one another.
What Is an LLM? (Large Language Model)
LLM (Large Language Model) is a type of AI specifically built to "understand" and "generate" human language. It is the technology behind ChatGPT, Claude, Gemini, and other conversational AI systems we are familiar with.
What Is an LLM Trained On?
LLMs learn from massive volumes of text data on the internet — websites, books, research papers, programming code, Wikipedia, and much more. The training data is measured in Terabytes (equivalent to millions of books).
How Does an LLM Work?
The fundamental principle of an LLM is "Predict Next Token" — selecting the most likely next word. When given text input, the LLM calculates the probability of what the next word should be and selects the one with the highest probability.
Example:
When given "The capital of Thailand is," the LLM calculates that the most probable next word is "Bangkok" because in the training data, these words appear together most frequently.
A Critical Limitation: Hallucination
Because LLMs work by "predicting" rather than "searching for truth," a problem called Hallucination arises — the AI "fabricates information" with confidence, such as citing non-existent research, providing incorrect statistics, or generating data that looks credible but is wrong. This is why you should never trust AI 100% without verification.
Essential Vocabulary for Executives
In AI meetings or technology articles, these terms appear frequently:
| Term | Meaning | Simple Analogy |
|---|---|---|
| Prompt Engineering | The art of crafting questions or instructions that help AI respond well and on point. Better prompts = better answers. | Like briefing an employee — the clearer the brief, the better the results |
| RAG (Retrieval-Augmented Generation) |
A technique that has AI read your organization's documents before responding, rather than answering from general knowledge. This reduces Hallucination and provides context-specific answers. | Like having an employee read the company manual before answering a customer |
| Fine-tuning | Customizing AI for specific domains by training it further with specialized data — such as medical, legal, or accounting data | Like sending an employee to a specialized training course |
| Token | The counting unit of AI — not words, but word fragments. English ~1 word = 1-2 Tokens. Thai ~1 word = 2-4 Tokens due to more complex characters. | Like phone call billing units (the more you talk, the more Tokens you use) |
| Temperature | The creativity level of AI. Low values (0.0-0.3) = straightforward, consistent answers. High values (0.7-1.0) = creative, varied, but potentially less accurate. | Like a dial adjusting AI's "willingness to think outside the box" |
| Hallucination | The phenomenon where AI fabricates information confidently as if it were factual. It occurs because AI "predicts" answers rather than "searching for" truth. | Like an employee who is afraid to say "I do not know" and guesses instead |
| Context Window | The amount of data AI can remember at once, measured in Tokens. For example, 128K Tokens means AI can read a document of ~200 A4 pages in one session. | Like desk size — the larger it is, the more documents you can spread out |
| Agent | AI that can execute multi-step tasks independently — planning, making decisions, using tools, and correcting errors, rather than just answering questions one by one. | Like an assistant you can delegate an entire project to, not just individual tasks |
How to Start Using AI in Your Organization — 4 Practical Steps
Many organizations want to start using AI but do not know where to begin. Here are 4 practical steps suited for Thai organizations:
Step 1: Start with Personal Experimentation (No Confidential Data)
Before deploying AI in the organization, executives and employees should try AI personally first to understand what it can and cannot do. Start with personal tasks unrelated to confidential organizational data, such as:
- Using AI to draft emails, summarize articles, translate documents
- Asking analytical questions like "Pros and cons of Cloud ERP vs On-premise ERP"
- Creating presentation outlines or brainstorming ideas
Cautions: Never enter confidential organizational data (financial statements, customer lists, passwords, contracts) into public AI — as that data may be used to further train the AI model.
Step 2: Identify Use Cases Where AI Truly Helps
Not every task benefits from AI. Identify tasks with these characteristics:
| AI Excels At | AI Is Not Yet Suitable For |
|---|---|
| Tasks requiring summarization of large data volumes | Tasks requiring 100% accuracy (e.g., accounting figures) |
| Drafting documents, emails, or initial reports | Tasks requiring legal or ethical decisions |
| Language translation and tone adjustment | Tasks involving confidential or personal data (when using public AI) |
| Brainstorming and analyzing alternatives | Tasks requiring real-time data references (stock prices, exchange rates) |
| Reviewing and editing documents | Tasks requiring genuine human empathy |
Step 3: Establish Your Organization's AI Policy
Before officially allowing employees to use AI, the organization should have clear policies:
- What data is off-limits for AI? Clearly define which confidentiality levels are prohibited from public AI use
- Who is responsible for verifying outputs? AI is not the final decision-maker; human review is always required
- Which AI tools are approved for use? Select AI tools with acceptable privacy policies
- Must AI usage be disclosed? Establish transparency policies for AI-assisted work
Step 4: Choose the Right AI for the Job
Each AI tool has different strengths:
| AI Type | Suitable for | Example |
|---|---|---|
| Conversational AI | Q&A, summarization, drafting, analysis | ChatGPT, Claude, Gemini |
| Image Generation AI | Graphic design, creating illustrations | Midjourney, DALL-E, Stable Diffusion |
| Coding AI | Assisting with writing and debugging code | GitHub Copilot, Claude Code, Cursor |
| Enterprise AI | Used with internal data, more secure | Azure OpenAI, AWS Bedrock, Self-hosted LLM |
For organizations with confidential data:
Consider using Enterprise AI where data never leaves your environment, or use RAG techniques that allow AI to read organizational documents in a secure environment without sending data for model training.
Precautions When Using AI
Although AI is highly beneficial, executives must understand these risks:
1. Hallucination (AI Fabrication)
As mentioned, AI can generate convincing but false information at any time — especially numbers, statistics, names, and citations. Always verify facts before using them, particularly for matters with business or legal implications.
2. Data Privacy
Data entered into public AI may be used to further train the model, meaning your organization's confidential information could be revealed in AI responses to other users. Each AI provider has different data policies — read them carefully.
3. Bias
AI learns from human-created data and therefore inherits biases — such as gender, racial, or cultural biases. When applying AI to tasks that affect people — like candidate screening or credit assessment — bias must be carefully monitored.
4. Hidden Costs
Using AI involves several hidden costs:
- API / Subscription fees: Enterprise-grade AI has significant monthly costs, especially at high usage volumes
- Learning time costs: Employees need time to learn how to use AI effectively
- Review costs: AI outputs require human review — workload is not reduced by 100%
- Infrastructure costs: Self-hosting AI requires additional server and maintenance expenses
Saeree ERP and AI
Currently, Saeree ERP does not yet have AI features built into the system. However, the development team is actively studying and planning to incorporate AI in the future, focusing on areas where AI excels — such as data analysis, report generation, and trend forecasting.
Saeree ERP's Approach:
- Prioritizing data accuracy over speed of AI adoption
- AI will only be added to areas that have been thoroughly tested
- Client confidential data will never be sent outside under any circumstances
- Follow updates in our AI article category
Conclusion
Generative AI is a technology that is transforming how we work on a massive scale. But to use it effectively in an organization, executives must understand both its potential and limitations — not just following trends, but with a clear plan, thoughtful policies, and a prepared team.
AI is not here to replace people — but people who know how to use AI will replace those who do not. The key is not rushing to adopt AI, but first "understanding" what AI can do, what it cannot do, and which tasks in your organization it is best suited for.
— Saeree ERP Team
If your organization would like to discuss integrating technology with your ERP system, you can schedule a demo or contact our advisory teamfor further discussion.
