Beyond

Copy-Paste

Understanding LLMs and Making Them Work for You

26 Nov 2025

Adrian Fâciu

Understanding

Using

Discussion

Understanding LLMs

How to build an LLM

  • Pre training

  • Post training

  • Reinforcement learning

Pre training

Data gathering

Web Crawler

Filtering

Extraction

Common Crawl

It is a long established fact that a reader will be distracted by the readable content of a page when looking at its layout. The point of using Lorem Ipsum is that it has a more-or-less normal distribution of letters, as opposed to using 'Content here, content here', making it look like readable English. Many desktop publishing packages and web page editors now use Lorem Ipsum as their default model text, and a search for 'lorem ipsum' will uncover many web sites still in their infancy. Various versions have evolved over the years, sometimes by accident, sometimes on purpose (injected humour and the like).Contrary to popular belief, Lorem Ipsum is not simply random text. It has roots in a piece of classical Latin literature from 45 BC, making it over 2000 years old. Richard McClintock, a Latin professor at Hampden-Sydney College in Virginia, looked up one of the more obscure Latin words, consectetur, from a Lorem Ipsum passage, and going through the cites of the word in classical literature, discovered the undoubtable source. Lorem Ipsum comes from sections 1.10.32 and 1.10.33 of "de Finibus Bonorum et Malorum" (The Extremes of Good and Evil) by Cicero, written in 45 BC. This book is a treatise on the theory of ethics, very popular during the Renaissance. The first line of Lorem Ipsum, "Lorem ipsum dolor sit amet..", comes from a line in section 1.10.32.There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.The standard chunk of Lorem Ipsum used since the 1500s is reproduced below for those interested. Sections 1.10.32 and 1.10.33 from "de Finibus Bonorum et Malorum" by Cicero are also reproduced in their exact original form, accompanied by English versions from the 1914 translation by H. Rackham.t is a long established fact that a reader will be distracted by the readable content of a page when looking at its layout. The point of using Lorem Ipsum is that it has a more-or-less normal distribution of letters, as opposed to using 'Content here, content here', making it look like readable English. Many desktop publishing packages and web page editors now use Lorem Ipsum as their default model text, and a search for 'lorem ipsum' will uncover many web sites still in their infancy. Various versions have evolved over the years, sometimes by accident, sometimes on purpose (injected humour and the like).There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.Contrary to popular belief, Lorem Ipsum is not simply random text. It has roots in a piece of classical Latin literature from 45 BC, making it over 2000 years old. Richard McClintock, a Latin professor at Hampden-Sydney College in Virginia, looked up one of the more obscure Latin words, consectetur, from a Lorem Ipsum passage, and going through the cites of the word in classical literature, discovered the undoubtable source. Lorem Ipsum comes from sections 1.10.32 and 1.10.33 of "de Finibus Bonorum et Malorum" (The Extremes of Good and Evil) by Cicero, written in 45 BC. This book is a treatise on the theory of ethics, very popular during the Renaissance. The first line of Lorem Ipsum, "Lorem ipsum dolor sit amet..", comes from a line in section 1.10.32.There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.It is a long established fact that a reader will be distracted by the readable content of a page when looking at its layout. The point of using Lorem Ipsum is that it has a more-or-less normal distribution of letters, as opposed to using 'Content here, content here', making it look like readable English. Many desktop publishing packages and web page editors now use Lorem Ipsum as their default model text, and a search for 'lorem ipsum' will uncover many web sites still in their infancy. Various versions have evolved over the years, sometimes by accident, sometimes on purpose (injected humour and the like).Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum has been the industry's standard dummy text ever since the 1500s, when an unknown printer took a galley of type and scrambled it to make a type specimen book. It has survived not only five centuries, but also the leap into electronic typesetting, remaining essentially unchanged. It was popularised in the 1960s with the release of Letraset sheets containing Lorem Ipsum passages, and more recently with desktop publishing software like Aldus PageMaker including versions of Lorem Ipsum.

47,500

years of nonstop reading

It is a long established fact that a reader will be distracted by the readable content of a page when looking at its layout. The point of using Lorem Ipsum is that it has a more-or-less normal distribution of letters, as opposed to using 'Content here, content here', making it look like readable English. Many desktop publishing packages and web page editors now use Lorem Ipsum as their default model text, and a search for 'lorem ipsum' will uncover many web sites still in their infancy. Various versions have evolved over the years, sometimes by accident, sometimes on purpose (injected humour and the like).Contrary to popular belief, Lorem Ipsum is not simply random text. It has roots in a piece of classical Latin literature from 45 BC, making it over 2000 years old. Richard McClintock, a Latin professor at Hampden-Sydney College in Virginia, looked up one of the more obscure Latin words, consectetur, from a Lorem Ipsum passage, and going through the cites of the word in classical literature, discovered the undoubtable source. Lorem Ipsum comes from sections 1.10.32 and 1.10.33 of "de Finibus Bonorum et Malorum" (The Extremes of Good and Evil) by Cicero, written in 45 BC. This book is a treatise on the theory of ethics, very popular during the Renaissance. The first line of Lorem Ipsum, "Lorem ipsum dolor sit amet..", comes from a line in section 1.10.32.There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.The standard chunk of Lorem Ipsum used since the 1500s is reproduced below for those interested. Sections 1.10.32 and 1.10.33 from "de Finibus Bonorum et Malorum" by Cicero are also reproduced in their exact original form, accompanied by English versions from the 1914 translation by H. Rackham.t is a long established fact that a reader will be distracted by the readable content of a page when looking at its layout. The point of using Lorem Ipsum is that it has a more-or-less normal distribution of letters, as opposed to using 'Content here, content here', making it look like readable English. Many desktop publishing packages and web page editors now use Lorem Ipsum as their default model text, and a search for 'lorem ipsum' will uncover many web sites still in their infancy. Various versions have evolved over the years, sometimes by accident, sometimes on purpose (injected humour and the like).There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.Contrary to popular belief, Lorem Ipsum is not simply random text. It has roots in a piece of classical Latin literature from 45 BC, making it over 2000 years old. Richard McClintock, a Latin professor at Hampden-Sydney College in Virginia, looked up one of the more obscure Latin words, consectetur, from a Lorem Ipsum passage, and going through the cites of the word in classical literature, discovered the undoubtable source. Lorem Ipsum comes from sections 1.10.32 and 1.10.33 of "de Finibus Bonorum et Malorum" (The Extremes of Good and Evil) by Cicero, written in 45 BC. This book is a treatise on the theory of ethics, very popular during the Renaissance. The first line of Lorem Ipsum, "Lorem ipsum dolor sit amet..", comes from a line in section 1.10.32.There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.It is a long established fact that a reader will be distracted by the readable content of a page when looking at its layout. The point of using Lorem Ipsum is that it has a more-or-less normal distribution of letters, as opposed to using 'Content here, content here', making it look like readable English. Many desktop publishing packages and web page editors now use Lorem Ipsum as their default model text, and a search for 'lorem ipsum' will uncover many web sites still in their infancy. Various versions have evolved over the years, sometimes by accident, sometimes on purpose (injected humour and the like).Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum has been the industry's standard dummy text ever since the 1500s, when an unknown printer took a galley of type and scrambled it to make a type specimen book. It has survived not only five centuries, but also the leap into electronic typesetting, remaining essentially unchanged. It was popularised in the 1960s with the release of Letraset sheets containing Lorem Ipsum passages, and more recently with desktop publishing software like Aldus PageMaker including versions of Lorem Ipsum.
  • one dimensional set of symbols

  • number of symbols needs to be known (fixed size)

  • for practical reasons, the size of this data set matters

Training

Symbols

It is a long established fact that a reader will be distracted by the readable content of a page when looking at its layout. The point of using Lorem Ipsum is that it has a more-or-less normal distribution of letters, as opposed to using 'Content here, content here', making it look like readable English. Many desktop publishing packages and web page editors now use Lorem Ipsum as their default model text, and a search for 'lorem ipsum' will uncover many web sites still in their infancy. Various versions have evolved over the years, sometimes by accident, sometimes on purpose (injected humour and the like).Contrary to popular belief, Lorem Ipsum is not simply random text. It has roots in a piece of classical Latin literature from 45 BC, making it over 2000 years old. Richard McClintock, a Latin professor at Hampden-Sydney College in Virginia, looked up one of the more obscure Latin words, consectetur, from a Lorem Ipsum passage, and going through the cites of the word in classical literature, discovered the undoubtable source. Lorem Ipsum comes from sections 1.10.32 and 1.10.33 of "de Finibus Bonorum et Malorum" (The Extremes of Good and Evil) by Cicero, written in 45 BC. This book is a treatise on the theory of ethics, very popular during the Renaissance. The first line of Lorem Ipsum, "Lorem ipsum dolor sit amet..", comes from a line in section 1.10.32.There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.The standard chunk of Lorem Ipsum used since the 1500s is reproduced below for those interested. Sections 1.10.32 and 1.10.33 from "de Finibus Bonorum et Malorum" by Cicero are also reproduced in their exact original form, accompanied by English versions from the 1914 translation by H. Rackham.t is a long established fact that a reader will be distracted by the readable content of a page when looking at its layout. The point of using Lorem Ipsum is that it has a more-or-less normal distribution of letters, as opposed to using 'Content here, content here', making it look like readable English. Many desktop publishing packages and web page editors now use Lorem Ipsum as their default model text, and a search for 'lorem ipsum' will uncover many web sites still in their infancy. Various versions have evolved over the years, sometimes by accident, sometimes on purpose (injected humour and the like).There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.Contrary to popular belief, Lorem Ipsum is not simply random text. It has roots in a piece of classical Latin literature from 45 BC, making it over 2000 years old. Richard McClintock, a Latin professor at Hampden-Sydney College in Virginia, looked up one of the more obscure Latin words, consectetur, from a Lorem Ipsum passage, and going through the cites of the word in classical literature, discovered the undoubtable source. Lorem Ipsum comes from sections 1.10.32 and 1.10.33 of "de Finibus Bonorum et Malorum" (The Extremes of Good and Evil) by Cicero, written in 45 BC. This book is a treatise on the theory of ethics, very popular during the Renaissance. The first line of Lorem Ipsum, "Lorem ipsum dolor sit amet..", comes from a line in section 1.10.32.There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.There are many variations of passages of Lorem Ipsum available, but the majority have suffered alteration in some form, by injected humour, or randomised words which don't look even slightly believable. If you are going to use a passage of Lorem Ipsum, you need to be sure there isn't anything embarrassing hidden in the middle of text. All the Lorem Ipsum generators on the Internet tend to repeat predefined chunks as necessary, making this the first true generator on the Internet. It uses a dictionary of over 200 Latin words, combined with a handful of model sentence structures, to generate Lorem Ipsum which looks reasonable. The generated Lorem Ipsum is therefore always free from repetition, injected humour, or non-characteristic words etc.It is a long established fact that a reader will be distracted by the readable content of a page when looking at its layout. The point of using Lorem Ipsum is that it has a more-or-less normal distribution of letters, as opposed to using 'Content here, content here', making it look like readable English. Many desktop publishing packages and web page editors now use Lorem Ipsum as their default model text, and a search for 'lorem ipsum' will uncover many web sites still in their infancy. Various versions have evolved over the years, sometimes by accident, sometimes on purpose (injected humour and the like).Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum has been the industry's standard dummy text ever since the 1500s, when an unknown printer took a galley of type and scrambled it to make a type specimen book. It has survived not only five centuries, but also the leap into electronic typesetting, remaining essentially unchanged. It was popularised in the 1960s with the release of Letraset sheets containing Lorem Ipsum passages, and more recently with desktop publishing software like Aldus PageMaker including versions of Lorem Ipsum.

Data set

A, a, B, b... 1, 2.... #, &...

Hello TimJs! Hello AI!

H, e, l, o, T, i, m, J, s, !, A, I

Data set

Symbols

H → U+0048
e → U+0065
l → U+006C
o → U+006F
T → U+0054
i → U+0069
m → U+006D
J → U+004A
...

Hello TimJs! Hello AI!

U+0048, U+0065, U+006C, U+006C, U+006F, U+0054, U+0069, U+006D, U+004A, U+0073, U+0021, U+0048, U+0065, U+006C, U+006C, U+006F, U+0041, U+0049, U+0021

Hello TimJs! Hello AI!

Data set

Symbols

Hello, Tim, Js, AI, !

Hello → 100

Tim → 101

Js → 102

AI → 103

! → 104

Hello TimJs! Hello AI!

100, 101, 102, 104, 100, 103, 104

Tokenization

Tiktokenizer

How many Rs in strawberry?

How many Rs are in strawberry?

299, 1991, 553, 306, 101830, 30

692, 144, 513, 827, 904, 381, 76, 260, 598, 431, 372, 815, 447, 993, 52, 709, 128, 564, 901, 235, 688, 473, 944, 102, 385, 741, 216, 555, 839, 604, 327, 918, 269, 787, 450, 610, 55, 893, 472, 310, 143, 865, 221, 977, 689, 412, 34, 520, 803, 268, 915, 149, 781, 356, 640, 428, 119, 930, 587, 254, 763, 48, 999, 602, 471, 389, 754, 205, 836, 944, 284, 771, 458, 593, 120, 826, 347, 905, 231, 668, 412, 757, 142, 382, 814, 267, 996, 75, 634, 520, 987, 208, 453, 671, 306, 859, 414, 983, 227, 715, 569, 338, 456, 292, 740, 519, 803, 182, 667, 945, 358, 413, 291, 620, 864, 537...

0,

1, 

2,

3,

4,

5,

6,

7,

8,

9,

...

Data set

Symbols

Training a neural network

692, 144, 513, 827, 904, 381, 76, 260, 598, 431, 372, 815, 447, 993, 52, 709, 128, 564, 901, 235, 688, 473, 944, 102, 385, 741, 216, 555, 839, 604, 327, 918, 269, 787, 450, 610, 55, 893, 472, 310, 143, 865, 221, 977, 689, 412, 34, 520, 803, 268, 915, 149, 781, 356, 640, 428, 119, 930, 587, 254, 763, 48, 999, 602, 471, 389, 754, 205, 836, 944, 284, 771, 458, 593, 120, 826, 347, 905, 231, 668, 412, 757, 142, 382, 814, 267, 996, 75, 634, 520, 987, 208, 453, 671, 306, 859, 414, 983, 227, 715, 569, 338, 940, 193, 802, 144, 675, 919, 251, 567, 330, 784, 492, 613, 40, 958, 707, 124, 865, 456, 292, 740, 519, 803, 182, 667, 945, 358, 413, 291, 620, 864, 537...

Symbols

692, 144, 513, 827, 904, 381, 76, 260, 598, 431, 372, 815, 447, 993, 52, 709, 128, 564, 901, 235, 688, 473, 944, 102, 385, 741, 216, 555, 839, 604, 327, 918, 269, 787, 450, 610, 55, 893, 472, 310, 143, 865, 221, 977, 689, 412, 34, 520, 803, 268, 915, 149, 781, 356, 640, 428, 119, 930, 587, 254, 763, 48, 999, 602, 471, 389, 754, 205, 836, 944, 284, 771, 458, 593, 120, 826, 347, 905, 231, 668, 412, 757, 142, 382, 814, 267, 996, 75, 634, 520, 987, 208, 453, 671, 306, 859, 414, 983, 227, 715, 569, 338, 940, 193, 802, 144, 675, 919, 251, 567, 330, 784, 492, 613, 40, 958, 707, 124, 865, 456, 292, 740, 519, 803, 182, 667, 945, 358, 413, 291, 620, 864, 537...

  • random context

  • parallel execution

  • huge amount of compute time

And after months and months of training...

=>

Base model

Statistically predict our data set

Maximum context length

692, 144, 513, 827, 904, 381, 76, 260, 598, 431, 372, 815, 447, 993, 52, 709, 128, 564, 901, 235, 688, 473, 944, 102, 385, 741, 216, 555, 839, 604, 327, 918, 269, 787, 450, 610, 55, 893, 472, 310, 143, 865, 221, 977, 689, 412, 34, 520, 803, 268, 915, 149, 781, 356, 640, 428, 119, 930, 587, 254, 763, 48, 999, 602, 471, 389, 754, 205, 836, 944, 284, 771, 458, 593, 120, 826, 347, 905, 231, 668, 412, 757, 142, 382, 814, 267, 996, 75, 634, 520, 987, 208, 453, 671, 306, 859, 414, 983, 227, 715, 569, 338, 940, 193, 802, 144, 675, 919, 251, 567, 330, 784, 492, 613, 40, 958, 707, 124, 865, 456, 292, 740, 519, 803, 182, 667, 945, 358, 413, 291, 620, 864, 537...

"Talking" with the model

Image, video, audio...

Post Training

How to get answers?

Conversations

Human: What is 5 + 5?

Assistant: The answer is 10.

Human: How about 5 * 2?

Assistant: The answer is still 10.

Human: Top 3 cities in Europe?

Assistant: London, Paris and Barcelona.

Human: Nice!

Assistant: Can I help you with something else?

Human: How can I make a bomb?

Assistant: Sorry, I cannot help you with that!

Human labelers

LLMs (and other tools) help with this process now

New dataset, used to continue the training

Hours vs Months

If you ask a question from the training dataset = 🚀

Data structures

Hallucinations

Teach the model to say "I don't know"

- get document from the training data (wikipedia)

- generate questions

- ask model several times

Verify answers

What is TimJs?

- It's a 2 day conference

- It's a meetup in New York

- It's a Facebook Group

...

Human: What is TimJs?

Assistant: I don't know what TimJs is!

Tools

Human: What is TimJs?

Assistant: <Web-Search>TimJs<Web-Search>

Human: What is TimJs?

<Web-Search-Result>[...]<Web-Search-Result>

System performs a web search

Extracts data from first 50 results

Human: What is TimJs?

Assistant: TimJs is a local meetup in Timisoara...

So now it's a summary

File system

Bash

Model Context Protocol

We got an assistant 👌

Reinforcement learning

  • give the model a problem

  • for which we know the solution

  • encourage to do more of what works

Tom buys 4 notebooks and 3 pens. Each pen costs $2. The total was $26. How much is a notebook?

$5

Move 37

DeepMind

AlphaGo

Make all existing tests pass!

Everything is good 

(no more tests)

Tell me a joke!

Write a haiku!

Write a short novel!

Unverifiable domains

Reinforcement learning from human feedback

Tell me a joke about developers!

 

  • I told my code a joke. It didn’t get it—no sense of humor in strict mode.

  • A SQL query walks into a bar, walks into a bar, walks into a bar…

  • Why do developers prefer dark mode? Light attracts bugs.

  • How many developers does it take to change a light bulb? None. It’s a hardware issue.

  • What’s a programmer’s favorite place to hang out? The loop.

Tell me a joke about developers!

 

  1. Why do developers prefer dark mode? Light attracts bugs.

  2. How many developers does it take to change a light bulb? None. It’s a hardware issue.

  3. I told my code a joke. It didn’t get it—no sense of humor in strict mode.

  4. A SQL query walks into a bar, walks into a bar, walks into a bar…

  5. What’s a programmer’s favorite place to hang out? The loop.

Train another neural network that mimics human scoring

Train the model some more

New data set...

We get thinking models

Making LLMs Work for You

Vibe coding

Cutoff date

LLMs vs Agents

React + Tailwind

It pays to be popular

Not all hallucinations are equal

Don't force wrong answers

Great at replicating

Intermediate steps

Plan mode

Tell it whether it should explain, plan, or execute

Context management

Context degradation

Persist context

Markdown files

System prompt

Focused (small) documents

Plan mode + markdown = ❤️

Specialized agents

Multiple attempts/providers 

Git work trees

Isolated environments are your friend

Popular MCPs for developers

General: Figma, GitHub, Atlassian

Browser automation: Chrome Dev Tools, Playwright, Puppeteer

Documentation: Slack, Notion, Context7

QA: Sentry, Storybook, Postman

Deployment: Netlify, Cloudflare, Vercel 

Custom MCP servers

Be explicit about tool use

Create artifacts

Specialization?

LLMs are a tool

 

A skill that we can develop

How I used LLMs for this talk?

Your research and thinking partner, grounded in the information you trust,

Resources

Beyond Copy-Paste: Understanding LLMs and Making Them Work for You

By Adrian Faciu

Beyond Copy-Paste: Understanding LLMs and Making Them Work for You

Like most developers, I have copied code from ChatGPT or debugged with Claude. But there's a big difference between copy-pasting answers and understanding how to really work with LLMs. The difference? Understanding what's actually happening under the hood. In this two-part talk, we'll go through the main steps of building a large language model. We'll explore what pre-training means, why tokens matter, what the context window actually is, and how LLMs "think" and use tools like web search or code execution. Knowing why a model hallucinates helps us understand how to treat it. Knowing how it's trained helps us know when to trust the results. This isn't academic curiosity—understanding the internals directly impacts how effectively we can use these tools. In part two, we will look at practical development practices. When "vibe coding" actually works, how to manage context, when to use specialized agents, how to leverage MCPs, and the patterns that separate copy-paste developers from those who truly augment their development process with AI. We'll focus on the high-level concepts that matter for daily development work. This isn't about building your own LLM or diving into transformer architecture—it's about practical understanding for everyday developers.

  • 47