LLMs - Large Language Models

LLMs are neural networks that can process and generate natural language text.

TRAINING PHASE

They are trained on a dataset of billions of sentences using unsupervised learning techniques. In the training process LLMs learn what is the most likely word to came next to the previous one based on huge amount of data.

INPUT BY USER

LLMs accept as input a text prompt by a user and in relation with it generate in output text, word by word (token by token).

GENERATION OF THE OUTPUT

The generation process consists in predicting the next word on the base of previously generated words. LLMs are trained in doing this without any consciousness which is a prerogative of the human mind.

Building a Simple Large Language Model

In this example we use as data the dystopian novel “Nineteen Eighty-Four – 1984” by English writer George Orwell, published on 1949.

Using the text of the novel as a data source, the following tables were produced. I show only a part of them:

Word/Token	Occurrences
the	6249
of	3309
a	2482
and	2326
to	2236
was	2213
He	1959
It	1864
in	1759
that	1457
had	1311
his	1079
you	1011
not	827
with	771
as	672
At	654
they	642
for	615
IS	614
but	611
be	608
on	604
were	583
there	559
Winston	526
him	512
i	495
which	443
s	439
one	426
or	424
…	…

Word/Token	Word Next	Score	Probability
of	the	743	0,01139
It	was	589	0,00903
in	the	574	0,00880
He	had	355	0,00544
he	was	273	0,00418
on	the	230	0,00352
was	a	225	0,00345
there	was	223	0,00342
to	the	212	0,00325
O	Brien	205	0,00314
to	be	203	0,00311
and	the	203	0,00311
had	been	202	0,00310
the	party	195	0,00299
at	the	183	0,00280
that	he	167	0,00256
from	the	161	0,00247
with	a	158	0,00242
did	not	148	0,00227
that	the	147	0,00225
of	a	145	0,00222
of	his	145	0,00222
out	of	142	0,00218
was	not	130	0,00199
with	the	127	0,00195
he	could	124	0,00190
it	is	124	0,00190
in	his	123	0,00188
in	a	122	0,00187
They	were	122	0,00187
seemed	to	115	0,00176
was	the	110	0,00169
could	not	109	0,00167
he	said	109	0,00167
the	same	103	0,00158
for	the	101	0,00155
by	the	95	0,00146
for	a	92	0,00141
into	the	92	0,00141
she	had	87	0,00133
as	though	82	0,00126
they	had	80	0,00123
that	it	80	0,00123
have	been	79	0,00121
and	a	78	0,00120
it	had	77	0,00118
The	other	76	0,00116
of	them	76	0,00116
to	him	75	0,00115
the	telescreen	75	0,00115
BIG	BROTHER	73	0,00112
…	…	…	…

This is a simple diagram to understand how the text is generated word by word.

For example, if I start with BIG, LLM will probably generate BROTHER, and continuing we can produce this sentence:

BIG	BROTHER	was	a	sort	of	the	thought	…
Probability	0,00112	0,0005	0,00345	0,00095	0,00104	0,00023	0,00087

ChatGPT is a LLM

By using “prompt” mechanism you can ask ChatGPT for what you want using the natural language.

But how ChatGPT “UNDERSTAND” text inserted by the user?

The text is transformed and each word represented by a code that computer can processed.

A way to represent individual words is Word2Vec technique in natural language processing (NLP), in which each word is represented by a vector (a set of numbers). This helped a computer to assign a meaning to the word.

Word2Vec stands for “words as vectors”. It means expressing each word in your text corpus in n-dimensional space. The word’s weight in each dimension defines it for the model.

The meaning of the words is based on the context defined by its neighboring words where they are associated.

A simple example of word representation using the Word2Vec approach in two-dimensional space.

Man = [1,4]

Woman = [1,3]

Manager = [4,2]

Actress = [4,1]

Manager	-	Man	+	Woman	=	Actress
[4,2]	-	[1,4]	+	[1,3]	=	[4,1]

In the following picture we have the graphic representation.

This is what happens when you sent some prompt to ChatGPT.

The text is converted and split in tokens;

[10,10], [10,31], [10,15], [14,44], [8,5], …

(you, are , an, ICT, specialist, with, a, lot, of, experience)

An algorithm (like ChatGPT) makes some prediction and output text word by word.

[10,10],…

(you,can,have, an, important, and, well-paid, job)

ChatGPT and PROMPT ENGINEERING

Let us now analyze some techniques to better exploit the potential of ChatGPT.

DIRECTIONAL PROMPTING

If you submit the same question to ChatGPT many times, you will likely receive different answers.

How can you use directional prompting in order to get more precise answer?

You have to give more information and to be more descriptive when you define a prompt. You have to give clear instruction. This will help the model to understand of what you want. If you ask for generic question, you receive generic answer.

Generic question:

BIBLIOGRAPHY/WEBOGRAPHY

[01] openai.com;

[02] KENNETH WARD CHURCH, Emerging Trends Word2Vec, IBM 2016;

Not generated by AI tools or platforms.

{[(homo scripsit)]}

"Artificial Intelligence attempts to coax a machine, typically a computer, to behave in ways humans judge to be intelligent" John McCarthy (1927-2011)

"I think that there is a lot of fear about robots and artificial intelligence among some people, whereas I'm more afraid of natural stupidity" Eugenia Cheng

Now we can talk with a “machine” using a natural language instead of using programming languages consisting of specific words to be written following a strict syntax and form.

We have a new paradigm in the HCI (Human Computer Interaction) with generative models and large language models.

We can interact with these AI Systems by using “prompt” mechanism, in which we have flexible inputs continued by equally flexible outputs.

In [01] this mechanism is called massive multimodal models.

In the following table a selection of different input/output modalities.

"Interaction with prompt-commanded AI is a different from other ways of interaction with machines" [01]. It has three important properties:

flexibility: using of text, code, images etc.;
generality: applicable to broad range of tasks;
originality: generate original content.

"Cognitive tools are external artifacts that are used to aid the psychological capacities of the human brain in completing a cognitive task" [01] They are used to reduce the cognitive work of human brain.

Massive multimodal models are cognitive tools or extenders they can be used for simple o complex interactions. The results depend on the skill and capacity of the user in exploiting these cognitive tools.

AI IMAGE SYNTHESIS: AI Text-to-Art Generator

It is the task where AI learns to understand a description in natural language and reproduce realistic image matching the description. It combines natural language processing (NLP) and computer vision (CV). In this text-to-image tasks NLP model is the encoder and an image synthesis model as the decoder.

Our society is becoming increasingly visual. Images are a very strong means of communication and in this Artificial Intelligence is a very powerful tool.

We can create incredible images (AI Text-to-Art Generator) using AI. Here are some websites where we can create image using AI:

www.bing.com/images/create: AI-powered Bing using its new feature Image Creator: “Powered by the very latest DALL∙E models from our partners at OpenAI, Bing Image Creator allows you to create an image simply by using your own words to describe the picture you want to see. Now users off the waitlist can generate both written and visual content in one place from within chat.”;
www.midjourney.com: you can access by discord.com account.
openai.com/dall-e-2: DALL·E 2 is an AI system that can create realistic images and art from a description in natural language. It is not free by default.
YOUIMAGINE from you.com: to magically transform your ideas into stunning visuals and one-of-a-kind graphics;
stable-diffusion-art.com: Stable Diffusion Art:
www.canva.com: MagicStudio By Canva allows to supercharge your work and designs with all power of AI.
www.imagine.art : "Create awe-inspiring masterpieces effortlessly and explore the endless possibilities of AI generated art".
davinci.ai : Create AI art using only your words in just a few seconds!

AI-powered Bing’s Image Creator

More specific is the prompt better you obtain the image that you have in mind. Bing’s Image Creator recommends you format your prompts:

Adjective + Noun + Verb + Style.

Small fox running in the forest, digital art

Simple prompt description:

“The centurion in the time of the Roman Empire: the backbone of the Roman army.”

in Bing's Image Creator produce in output:

Complex prompt description

"[...] the centurions must be, not so much men who are bold and contemptuous of danger, as men who are able to command, tenacious and calm, who, moreover, do not move to attack when the situation is uncertain, nor throw themselves into the heat of battle, but on the contrary know how to resist even when pressed and defeated, and are ready to die on the battlefield." POLYBIUS, HISTORIES, VI, 24, 9

Produce:

MIDJOURNEY

Let's have a look to midjourney: an AI image generator prompt. A prompt is an input that guides a computer’s AI system in producing an art.

Prompts can range from a simple text description:

“The centurion in the time of the Roman Empire: the backbone of the Roman army.”

that produce:

to more complicated description that involve multiple parts coming together:

Produce:

In midjourney the syntax of the prompt in order to generate an image is the following:

/imagine < description text of the image >

What you put in the prompt is very important in order to define the picture that you would have be generated by midjourney.com.

You can ask :

a photo of ...txt...

a painting of ...txt...

you can decide the subject of the photo or painting:

animal;
person;
landscape;
object;
and so on

You can define which details you would like to add:

1) special environment: for example on a boat, in the forest,...

2) special lighting:

soft lighting,
ring lighting,
neon,
and son on

3) colour scheme;

4) point of view:

camera behind;
camera in the front of;
camera beside;

5) background:

solid colour;
a nebula;
a forest;
and so on.

6) atmosphere:

vibrant;
dark;
and so on.

You can add more information, for example the time of the day and so on.

As an image is 1000 words you can ask midjourney.com to generate an image by an uploaded image.

You can merge multiple images into one by using the blending process.

/blend <image1> <image2> ...

You can add additional text to enrich or modify the image.

You can ask also midjourney to get the prompt back (image captioning) using this command syntax:

/describe <image>

This means please describe this image for me.

You can use a negative prompting about you don't really want in the results, putting at the end of your prompt:

--no fog or dust

The you can use other commands:

--aspect 5:4 for aspect ratio

--ar 5:4

--chaos 100 or 90

You can stop the process at a determined percent in order to have an image at different stages.

--stop 50

Human-Computer Interaction but Human-Centred

Massive multimodal models are cognitive extenders and are distinct from autonomous AI systems because they are highly user-dependent.

BIBLIOGRAPHY/WEBOGRAPHY

[01] Wout Schellaert et al., Your Prompt is My Command, Journal of Artificial Intelligence Research, 2023;

[02] Ronald T. Kneusel, How AI works, 2024;

Not generated by AI tools or platforms.

{[(homo scripsit)]}

SCIENTIFIC THOUGHTS IN ACTION

Natural Language Processing (NLP)

AI in Action: BASIC CONCEPTS ABOUT LLM

Building a Simple Large Language Model

ChatGPT is a LLM

ChatGPT and PROMPT ENGINEERING

BIBLIOGRAPHY/WEBOGRAPHY

AI IN ACTION: COOPERATION WITH HUMAN

AI IMAGE SYNTHESIS: AI Text-to-Art Generator

AI-powered Bing’s Image Creator

MIDJOURNEY

Human-Computer Interaction but Human-Centred

BIBLIOGRAPHY/WEBOGRAPHY