Demystifying LLMs- How to Use GenAI Effectively.pdf

Large Language Models (LLMs) like ChatGPT and Gemini are reshaping how people in every field approach problem-solving, from brainstorming ideas to debugging complex systems. This talk will explain, in straightforward terms, how LLMs work, what they’re good at, and where their boundaries lie.
We’ll explore the principles of prompt engineering—how to frame questions and instructions to get clearer, more reliable results—and show practical ways to make LLMs a valuable part of everyday work. Attendees will leave with a grounded understanding of how to think about LLMs and concrete strategies for using them effectively in their own workflows.

1. Demystifying LLMs: How to
Use GenAI Effectively

to keep your job


Aiko Klostermann

2. What are LLMs?
LLMs are sophisticated mathematical functions that predict what word comes next for any piece of text

3. How do LLMS work?
•Deep learning systems trained to recognize and generalize patterns
•Training: Shown trillions of tokens → predicts next token → sees the real one → updates weights via
backpropagation
•Learns broad structures of language and statistical relationships between tokens - not to memorized facts
•LLMs specialize in sequence modeling
–They predict the next token given all previous context
–Tokens → embeddings: words are mapped into a high-dimensional space where similar meanings cluster
together.
–Token-by-token generation
•The model predicts one token, then reprocesses the entire context (your prompt + all previous
tokens) to predict the next one, repeating until completion.

4. Multiply two 3 digit prime numbers.
The response format is:
Result: <product>
Factor 1: <factor 1>
Factor 2: <factor 2>
Multiply two 3 digit prime numbers.
The response format is:
Factor 1: <factor 1>
Factor 2: <factor 2>
Result: <product>
What can they do? 

5. What can they do?
Demo

6. What did we learn from that?
Token prediction
The first example fail since
the prime numbers were
not yet in the context
02.
Prompt engineering
Prompt engineering can
set the right context
03.
Correctness
LLMs will confidently tell
you something incorrect.
01.

7. shifting the
statistical space
the model draws
from
Persona
Intent
Output format
“You are chatGPT, a large language model…”
“You are a Senior Software Engineer…”
“Generate Python script”
Provide output examples
Prompt engineering - Set Context
LLMs don’t infer goals, state what you’re trying to achieve
“Summarize this email for a busy executive”

8. Communication
Design
discussions
Documentation
Explaining systems
Knowledge retrieval
Debugging
Draft technical emails,
PRDs, summarise meeting
notes, presentations …
Draft README files,
internal docs …
Brainstorm API design, UI
Prototype, system design
patterns …
Complexity in plain English
conversational index for
frameworks, libraries …
Interpret stack traces,
configuration errors …
Use
Cases

9. Thank you!
Aiko @
aiko.dev
open.gov.sg
name websites
email addresses