Introduction
Everyone around you is talking about AI. Your manager wants an “AI strategy.” Your vendor is calling everything “AI-powered.” And Oracle just released version 23ai — named after Artificial Intelligence itself.
But if you’ve spent your career mastering SQL, RMAN, Data Guard and PL/SQL, you might be staring at terms like “LLM,” “embeddings” and “vector search” and thinking: where do I even start?
This post is your starting point. No math. No Python. No assumptions about what you already know. By the end, you will be able to clearly explain what AI is, what Machine Learning actually means, what a Large Language Model does, and — most importantly — why all of this is now directly relevant to you as an Oracle DBA.
What you’ll learn:
- The difference between AI, Machine Learning and Deep Learning — in one clear diagram
- What a Large Language Model (LLM) is and how it actually works, without equations
- Why Oracle added “ai” to its version number and what that means for your career
- The three questions every DBA should be asking about AI right now
Prerequisites
This is a concept post — no software required.
- You are comfortable working with Oracle databases (any version)
- You have heard terms like “AI,” “ChatGPT” or “LLM” but want a proper grounding
- You have about 15 minutes to read carefully
The Problem With How AI Is Usually Explained
Most AI explanations go one of two ways.
The first way is too abstract: “AI is a simulation of human intelligence in machines.” That sentence sounds meaningful but tells you nothing you can use.
The second way is too mathematical: activation functions, gradient descent, backpropagation. Useful eventually — but not on day one.
There is a third way: use analogies that connect to what you already know as a DBA. That is exactly what this post does.
Section 1: AI, ML and Deep Learning — The Russian Dolls
Think of these three terms as nested containers — like Russian dolls. Each one sits inside the next.
┌─────────────────────────────────────────────┐
│ │
│ ARTIFICIAL INTELLIGENCE (the biggest) │
│ │
│ ┌─────────────────────────────────────┐ │
│ │ │ │
│ │ MACHINE LEARNING (inside AI) │ │
│ │ │ │
│ │ ┌─────────────────────────────┐ │ │
│ │ │ │ │ │
│ │ │ DEEP LEARNING │ │ │
│ │ │ (inside ML) │ │ │
│ │ │ │ │ │
│ │ │ ┌───────────────────────┐ │ │ │
│ │ │ │ │ │ │ │
│ │ │ │ LLMs live here │ │ │ │
│ │ │ │ (ChatGPT, Claude, │ │ │ │
│ │ │ │ Llama, Cohere...) │ │ │ │
│ │ │ │ │ │ │ │
│ │ │ └───────────────────────┘ │ │ │
│ │ └─────────────────────────────┘ │ │
│ └─────────────────────────────────────┘ │
└─────────────────────────────────────────────┘
Let’s open each one.
Artificial Intelligence — The Outer Shell
Artificial Intelligence is any technique that makes a computer do something that normally requires human intelligence.
The key word is normally. Chess programs were called “AI” in 1997. Spam filters are AI. So is the autocomplete on your phone. AI is not one specific thing — it is a broad category of “computers doing smart things.”
As an Oracle DBA, you have probably already used AI without calling it that. Oracle’s Automatic Workload Repository (AWR) and Automatic Database Diagnostic Monitor (ADDM) analyse performance patterns and make recommendations. That is, technically, a form of AI.
Machine Learning — The Middle Layer
Machine Learning is a specific approach to building AI. Instead of writing rules by hand (“if query takes more than 5 seconds, alert“), you feed the computer thousands of examples and let it figure out the patterns itself.
Here is the DBA analogy: imagine you wanted to predict whether a query would cause a full table scan. The old approach is to write IF no_index_on_column THEN alert. Machine Learning says: give me 50,000 historical query execution plans and their outcomes, and I will figure out the pattern myself — including patterns you never thought to write rules for.
The result is a model — a mathematical summary of everything the computer learned from those examples. When a new query comes in, the model predicts what will happen, based on what it learned before.
This is fundamentally different from traditional programming:
| Traditional Programming | Machine Learning |
|---|---|
| You write the rules | The computer learns the rules |
| Rules are explicit and readable | Rules are implicit, inside the model |
| Adding new cases requires new code | Retrain with new data |
| Predictable and auditable | Probabilistic — not always right |
| Good when rules are known | Good when rules are too complex to write |
Deep Learning — The Inner Layer
Deep Learning is a specific type of Machine Learning that uses structures loosely inspired by the human brain, called neural networks.
The “deep” part refers to the many layers in these networks. Each layer transforms data a little more — detecting simple patterns at first, then more complex ones, then extremely abstract ones.
Deep Learning is why computers can now:
- Recognise faces in photos
- Transcribe speech to text
- Translate between languages
- Generate realistic images from text descriptions
LLMs — Large Language Models — are a specific type of deep learning model trained on text. Which brings us to the most important concept in this series.
Section 2: What Is a Large Language Model?
A Large Language Model (LLM) is a deep learning model trained on an enormous amount of text — web pages, books, code, documentation, scientific papers — with one original goal: predict the next word in a sentence.
That sounds deceptively simple. Let us look at it carefully.
The Training Process (Simplified)
Imagine feeding the model this text:
“The Oracle database stores data in…”
The model’s job during training is to predict what word comes next. It guesses. It gets told whether it was right. It adjusts. Then it sees the next sentence. And the next. And the next.
It does this billions of times, across trillions of words.
After all that training, something remarkable happens. To predict the next word accurately in any context — technical, legal, medical, conversational — the model has to develop a deep internal understanding of:
- Grammar and syntax
- Facts about the world
- Logic and reasoning
- Relationships between concepts
- How to write in different styles and tones
The model never explicitly “memorises” a Wikipedia article. Instead, it develops a compressed internal representation of knowledge — stored as billions of numerical parameters (the model’s “weights”).
Why “Large”?
The “Large” in LLM refers to two things:
- The size of the training data — GPT-4 was trained on roughly 45 terabytes of text. For reference, the entire English Wikipedia is about 20 gigabytes.
- The number of parameters — parameters are the numbers inside the model that store what it learned. GPT-3 had 175 billion parameters. Larger models generally perform better, but cost more to run.
How an LLM Actually Responds to You
When you type a question into ChatGPT or Claude, here is what happens:
Your Input
│
▼
┌─────────────────────────────────┐
│ TOKENISATION │
│ Text is broken into tokens │
│ "Oracle" → one token │
│ "database" → one token │
│ "23ai" → possibly two tokens │
└───────────────┬─────────────────┘
│
▼
┌─────────────────────────────────┐
│ CONTEXT WINDOW │
│ All tokens fed into model │
│ Model "sees" your full prompt │
└───────────────┬─────────────────┘
│
▼
┌─────────────────────────────────┐
│ NEXT TOKEN PREDICTION │
│ Model predicts most likely │
│ next token, then the next, │
│ then the next... one at a time │
└───────────────┬─────────────────┘
│
▼
Your Response
(generated token by token)
The response is built one token at a time, each new token influenced by everything that came before it. This is why LLMs can write coherently across long passages — every word is chosen in context.
What LLMs Are Good At
| Task | Example | Quality |
|---|---|---|
| Explaining concepts | “Explain Oracle Data Guard to a junior DBA” | Excellent |
| Writing and editing | Draft a runbook, rewrite documentation | Excellent |
| Code generation | Write a PL/SQL procedure for X | Very Good |
| Summarising text | Summarise this AWR report | Very Good |
| Question answering | From a provided document | Good |
| Reasoning through problems | Debug this error step by step | Good |
What LLMs Are NOT Good At (Without Help)
This is crucial — and it is exactly why Oracle 23ai matters.
| Limitation | What It Means |
|---|---|
| Knowledge cutoff | LLMs are trained up to a point in time. They do not know what happened last week. |
| No access to your data | ChatGPT has never seen your Oracle database, your schema, your company documents. |
| Hallucination | LLMs sometimes generate confident, fluent, completely wrong answers. They cannot fact-check themselves. |
| No memory by default | Each conversation starts fresh. The model has no memory of last Tuesday’s conversation. |
Do you see the pattern? All four limitations are data problems. The LLM is powerful but needs your data to be truly useful in an enterprise context.
This is precisely the gap that Oracle 23ai is designed to fill.
Section 3: Popular LLMs You Should Know
You will encounter these names frequently. Here is a quick reference:
| LLM | Made By | Key Characteristic | Relevant to Oracle |
|---|---|---|---|
| GPT-4 / GPT-4o | OpenAI | Most widely used, very capable | Oracle Select AI can use it |
| Claude 3.5 / Claude Sonnet | Anthropic | Strong reasoning, large context | Oracle DBMS_VECTOR_CHAIN supports it |
| Llama 3 | Meta | Open source, can run on your own server | OCI hosts Llama via Gen AI service |
| Cohere Command | Cohere | Enterprise-focused, strong for RAG | Native in OCI Generative AI service |
| Mistral | Mistral AI | Efficient, open weights | Available on OCI |
You do not need to understand the internal differences yet. What matters for now: these are all available to call from inside Oracle 23ai, without leaving your database. We will show you exactly how in later posts.
Section 4: Why This Is a DBA Problem Now (Not Just a Developer Problem)
Here is a conversation happening in thousands of organisations right now:
Developer: “We want to build an AI chatbot that answers questions about our product data.”
Manager: “Great. Can it access our Oracle database?”
Developer: “LLMs don’t connect to Oracle directly. We’d need to build a pipeline to extract the data, convert it to vectors, store it somewhere the LLM can search, then query it and feed results back into the LLM.”
DBA (you): “…wait, you want to extract data out of Oracle to put it in some other system?”
This conversation reveals something important. The “AI stack” as most developers imagine it lives outside the database. Data gets exported. Processed. Stored in a specialist vector database. Queried. Fed into an LLM. Results come back.
Every one of those steps introduces:
- Data movement risk
- Latency
- Synchronisation problems
- Additional infrastructure to manage
- New security surfaces
- Licensing costs
Oracle’s answer with 23ai is: do it all inside the database you already have.
Store vectors natively. Generate embeddings with a SQL function. Call LLMs with PL/SQL. Run semantic search with a SQL query. The Oracle DBA who understands this is no longer just maintaining a database — they are building AI applications.
Section 5: Three Questions Every Oracle DBA Should Be Asking Right Now
Question 1: “What data do we have that an LLM would find useful?”
Think about your Oracle databases. What lives in there that people frequently ask questions about?
- Product catalogues
- Customer records and history
- Support tickets and resolutions
- Internal documentation stored as CLOBs
- Financial transaction records
Any of that could power an AI assistant. The data is already in Oracle. Oracle 23ai lets you build the AI layer on top of it — without moving the data.
Question 2: “What does ‘vector’ mean for something in our database?”
We will go deep on this in the next post. For now, understand this: a vector is just a way of representing any piece of information — a product description, a support ticket, a customer review — as a series of numbers that captures its meaning. LLMs can compare those numbers to find content that is similar in meaning, not just similar in keywords.
This is fundamentally different from a LIKE query or a full-text index.
Question 3: “What will my role look like in 2 years?”
DBAs who understand how to integrate LLMs into Oracle pipelines are already becoming rare and valuable. The skills are not hard — they build directly on SQL and PL/SQL knowledge you already have. But they require learning the new concepts first.
That is exactly what this series is built for.
Key Takeaways
✅ AI is a broad term — Machine Learning and Deep Learning are specific approaches within AI. LLMs are a specific type of Deep Learning model.
✅ LLMs predict the next token — Their remarkable capabilities (reasoning, coding, summarising) all emerge from this one training task applied at massive scale.
✅ LLMs have four core limitations — No knowledge of recent events, no access to your data, tendency to hallucinate, and no persistent memory. All are data problems.
✅ Oracle 23ai addresses all four limitations — By bringing vector storage, embedding generation and LLM calling inside the database itself.
✅ This is a DBA opportunity, not just a developer topic — The Oracle DBA who understands AI integration is positioned at the intersection of the most valuable data (in Oracle) and the most powerful new technology (LLMs).
Test Your Knowledge
You have just built your mental model of AI, ML and LLMs. Now test it:
👉 AI Basics for Oracle DBAs Quiz → gradeupnow.in/qsm_quiz/oracle-ai-basics-quiz/
20 questions · Covers AI vs ML vs DL, how LLMs work, LLM limitations · Instant feedback · Free
What’s Next
This post is part of the Oracle 23ai + LLM Integration series — taking you from zero to building real AI applications on OCI.
| # | Post | Status |
|---|---|---|
| 1 | What is AI, ML and LLM? — Plain English for Oracle DBAs | 📍 You are here |
| 2 | What is a Vector? — The Idea Behind AI Explained in 10 Minutes | ⬜ Coming next week |
| 3 | Why Do We Need a Vector Database? — The Problem Traditional DBs Can’t Solve | ⬜ Coming soon |
| 4 | Oracle 23ai Overview — Why Oracle Added AI to Its Database | ⬜ Coming soon |
👉 Next Post: What is a Vector? → gradeupnow.in/oracle-blog/what-is-vector-explained/
References
- Oracle 23ai Official Documentation — AI Features
- Oracle AI Vector Search Developer Guide
- OCI Generative AI Service Overview
- What are Large Language Models — Google Cloud Explanation
Found this helpful? Share it with your Oracle DBA team — they need this context too.
Questions or corrections? Drop them in the comments below. I read every one.