Why I am not afraid AI will take my job (Devoxx BE)

Large language models (LLMs) are here to stay. So is their impact on many professions. Our work in IT is no exception to that rule. Some doomsayers predict that such models will soon make large parts of our work redundant.

So, is it time to start queuing up in the unemployment line? I don’t think so! To understand why, I will explain how an LLM works behind the facade of the prompt window. Once we understand what an LLM essentially is, we will better understand the threats and the opportunities for our work.

Join me to get a better understanding of the inner workings of an LLM. As a bonus, you will leave the room with a much more optimistic view of what their impact on your work can be.

Transcript

1. And they lived… This sentence is commonly used to mark the end of many fairy tales.
3. Generative Pre-trained Transformer
4. LLM, AI, ML… Large Language Models Generative AI Deep Learning Machine Learning Artificial Intelligence
5. Aside: on the “IQ” of a LLM
6. Embedding Multi-Layer Perceptrons Attention Unembedding So, what is a Large Language Model? ~96 repeats
7. “Given a piece of text, what is the most likely next chunk of text?” LLMs excel in answering a simple question
8. Java is undoubtedly the... 60.26% 33.29% 2.41% 2.32% 1.72% 0.00% 0.00% 0.00% 0.00% 10.00% 20.00% 30.00% 40.00% 50.00% 60.00% 70.00% 80.00% 90.00% 100.00% language next world biggest fastest first best most
9. Java is undoubtedly the most... 9 86.31% 5.33% 4.58% 2.86% 0.01% 0.00% 0.00% 0.00% 0.00% 10.00% 20.00% 30.00% 40.00% 50.00% 60.00% 70.00% 80.00% 90.00% 100.00% common interesting well advanced important widely powerful popular
10. Java is undoubtedly the most popular... 10 49.64% 33.78% 5.78% 5.52% 5.28% 0.00% 0.00% 0.00% 0.00% 10.00% 20.00% 30.00% 40.00% 50.00% 60.00% 70.00% 80.00% 90.00% 100.00% framework open and browser operating platform programming language
11. Java is undoubtedly the most popular programming... 11 99.91% 0.01% 0.01% 0.01% 0.01% 0.00% 0.00% 0.00% 0.00% 10.00% 20.00% 30.00% 40.00% 50.00% 60.00% 70.00% 80.00% 90.00% 100.00% paradigm library programming framework platform environment languages language
13. 2. Embedding 1. Tokenization Embedding – From text to numbers 13
14. Vectors – and why they matter
15. Vectors capture meaning
17. Attention block ~ relates tokens to each other As water flows through the river, the vegetation on its bank…
18. Ex.: resolve ambiguity with earlier nouns water, river, vegetation
20. Multi-layer Perceptrons ~ link to existing knowledge Vocabulary that scares the shit out of you
21. User Prompt TL;DR Large Vocabulary Next Piece of Text System Prompt Alas, non- deterministic!
22. What is our job all about?
23. So, we don’t need AI? 23
24. Context
25. Ideation
26. Stakeholder Management (1)
27. Stakeholder Management (2)
28. Solution Strategy
29. User Stories
30. Screen Mockups & Specifications
31. Coding – the “boring” part
32. … and repeat!
33. So, how did AI change my job? ➢ Less “non-essential” work – esp. writing memos for managers! ➢ Less repetitive work and writing boiler-plate code! ➢ More attention to the details that matter!
34. Two key skills to learn (or improve) ➢ Natural language command ➢ Context engineering
35. AI might actually give useful output AI is no magical device solving all problems with enough context and precise prompts Lessons Learned Always keep humans in the loop
36. References & Thanks 36 • The “Deep Learning” series by Grant Sanderson (3Blue1Brown) on YouTube Transformers, the tech behind LLMs (chapter 5) Attention in transformers, step-by-step (chapter 6) How might LLMs store facts (chapter 7) • LLMs can't optimize schedules, but AI can! by Tom Cools • Unveiling the simple math that powers AI by Marianne Hoornenborg • A GitHub repository with context engineering examples for Claude Code and GitHub Copilot • How I Levelled Up My GitHub Copilot Prompts with Instruction Files and Context Engineering
37. Thank You! @mthmulders https://mthmulders.nl/