Decoding LLMs like ChatGPT: The 9 Unfiltered Factors

One day, ChatGPT gives you an answer that is a great goal, right in the corner. The next day, what he answers you is a 💩. If you're on that emotional rollercoaster called artificial intelligence, where one second you're king of the world and the next, you're looking at the instruction manual wondering where the catch is, this article is for you.

LLMs such as ChatGPT, that marvel of the technology that promises to make us the easier life, sometimes it seems that gets up on the wrong foot. But it's not a matter of luck; there's a science behind the whole mess. And yes, by understanding those gears, you can go from being a simple user to a master using this tool.

Here we are going to take a look at the ChatGPT engine.

Contrary to what is said of things like «to drive a car you don't need to know how a car's engine works«. In the case of IA, while anyone can use language models such as ChatGPT, in my opinion, yes, you need to know what it's about, if you want to be a PRO like Michael Schumacher driving.

Imagine you are trying to converse with someone from another culture or even another time without having any idea of their customs or language. If you don't know what their world is about, your conversation will probably leave you that confused.

Here are 9 crucial factors that decide how ChatGPT will respond to you.

So make yourself comfortable, because you're about to enter the heart of the matter, that place where data becomes words and words, hopefully, into something resembling an excellent coherent response.

1️⃣ Machine Learning and Neural Network Training: Key to ChatGPT

How does ChatGPT Learn?

Simplified explanation of how machine learning works:

Imagine you want to teach a child to recognise different types of animals.

1. You show the child pictures of different animals: Dogs, cats, horses, cows, etc.

2. You tell the child the name of each animal: «This is a dog», «This is a cat», etc.

3. You repeat this process many times: Over time, the child learns to recognise the different animals on its own.

This is a simple example of machine learning:

• The child is the model for machine learning.
• Animal images are the training data.
• The names of the animals are the DATA LABELS.
• The process of showing images and labels to the child is the training process.

In the case of neural networks:

• The machine learning model consists of a network of artificial neurons.
• The artificial neurons simulate, and are inspired by, neurons in the human brain.
• The artificial neurons connect with each other and form a network.
• The neural network learns to perform a task from training data.

In the case of ChatGPT:

• The training data is a large amount of text and code.
• The neural network analyses this data and learns to generate text similar to what it has seen in the dataset.
• ChatGPT can then use this knowledge to generate answers to your questions, write different types of creative content and translate languages.

2️⃣ Transformer Architecture: The Rebellion against Old School AI

The Coup d'Etat in the World of Language Processing

It is an innovative structure that allows ChatGPT to process and generate text efficiently and accurately.

Transformer architecture. No, we're not talking about those giant robots that transform into sports cars or trucks. We're into something much more impressive (and useful, unless you have to save the world from an alien invasion). Transformer architecture is that bit of technological genius that took the boring world of language processing and turned it on its head.

How does it work?

The Transformer architecture breaks text into small units called «tokens». It then analyses the relationships between these tokens to understand the meaning of the text as a whole.

This allows ChatGPT:

• Process text faster and more efficiently.
• Generate more coherent and contextually correct text.
• Better understand the relationships between words and sentences.

3️⃣ Attention mechanism: Focusing on what matters

Imagine you are at a party with lots of people talking.

To keep a conversation going, you need to focus on the person speaking and filter noise from the rest.

ChatGPT's care mechanism works in a similar way.

What is the attention mechanism?

It is a technique that allows ChatGPT to focus on the most relevant parts of your request to generate more accurate and relevant responses.

The attention mechanism is like a spotlight that illuminates the most important parts of a scene.

In conversation:

• Your brain acts as the focus, focusing on the person speaking.
• You filter out background noise, such as other people's conversations.
• You concentrate on the speaker's words, tone of voice and body language to understand what he or she is saying.

In the same way:

• ChatGPT uses the attention mechanism to focus on the keywords in your request.
• Assign greater importance to the words that are more relevant to the task you want to perform.
• Generate an accurate and relevant response based on the most important parts of your request.

Example:

When you ask a random LLMs like ChatGPT, «What is the capital of France, and what can you tell me about the Eiffel Tower?, the attention mechanism allows ChatGPT to discern that there are two parts to your question. First, it identifies the direct answer (Paris) and then focuses on providing specific information about the Eiffel Tower, rather than rambling on about other monuments or aspects of Paris.

4️⃣ Natural Language Processing (NLP): ChatGPT Understands, Not Just Listens

The Art of Understanding the Human Blah Blah Blah

What is NLP?

This is the branch of artificial intelligence that is responsible for understanding and interpreting human language.

How does it work?

NLP uses machine learning techniques to analyse large amounts of text data, identify patterns and learn to interpret the meaning of words and sentences.

What role does it play in ChatGPT?

It is essential for ChatGPT to understand your requests, generate accurate and relevant responses, and adapt its communication to different styles and contexts.

Examples of how NLP has improved ChatGPT:

• Sentiment analysis: ChatGPT can detect the emotion behind a request, be it joy, sadness, anger or frustration, and tailor its response accordingly.
• Machine translation: ChatGPT can translate texts from one language to another with greater accuracy and fluency thanks to advances in NLP.
• Creative text generation: ChatGPT can write poems, stories, scripts and other types of content. creative with greater originality and quality thanks to NLP techniques.

DISCLAIMER: While language models such as ChatGPT have no sentiment, If they can «simulate» understanding of feelings, they can «simulate» empathy, so much so that it seems very human. It really is an amazing technology, you can try asking them more emotional questions and asking for advice and you can test their «human side».

5️⃣ Self-supervised learning: The road to independence

Imagine a child learning to walk.

You don't need anyone to tell you how to do it. Just watch others walk, experiment on your own and, in time, master the skill.

The self-supervised learning ChatGPT works in a similar way.

What is self-supervised learning?

It is a method of learning where ChatGPT learns by itself from large amounts of data without the need for human intervention.

How does it work?

ChatGPT is fed with a large amount of text data, such as books, articles and conversations. From this data, it learns to identify patterns and relationships between words.

This allows you to:

• Improve their ability to produce coherent and grammatically correct text.
• Adapt their communication style to different contexts and situations.
• Better understand human emotions and feelings.

6️⃣ Tokenisation: Language in pieces

Imagine a puzzle huge.

To assemble it, you need to break it down into smaller, more manageable pieces.

Tokenisation works in a similar way.

What is tokenisation?

It is the process of dividing the language into small units called tokens. ChatGPT uses the tokenisation to process language more efficiently.

How does it work?

ChatGPT splits text into tokens of different types, such as words, punctuation and symbols.

This allows you to:

• Understand the structure of the text.
• Identify relationships between words.
• Generate new text that is coherent and grammatically correct.

Benefits of tokenisation:

• Efficiency: It allows ChatGPT to process language faster and more accurately.
• Accuracy: It reduces the possibility of errors in language comprehension.
• Flexibility: It allows ChatGPT to adapt to different types of language.

7️⃣ Semantic layers: Intelligence in action

Imagine a chef preparing a delicious dish.

He not only mixes ingredients, but also understands the properties of each ingredient to create a unique culinary experience.

ChatGPT's semantic layers work in a similar way.

What are semantic layers?

These are different levels of language comprehension that enable ChatGPT to understand the meaning of words in different contexts.

How do they work?

ChatGPT not only analyses individual words, but also relates them to other words, phrases and sentences.

This allows you to:

• Understand the meaning of the text at a deeper level.
• Identify the user's intentions.
• Generate more accurate and coherent responses.

Benefits of semantic layers:

• Accuracy: It reduces the possibility of errors in language comprehension.
• Coherence: It allows ChatGPT to generate responses that fit the context of the conversation.
• Naturalness: It allows ChatGPT to generate responses that sound as if they were written by a person.

Example:

The prayer «What time is it?»can have different meanings depending on the context.

• If you ask someone who is awake, it may mean that they want to know the current time.
• If you ask someone who is asleep, it may mean that they want to know if it is time to wake up.

Here are some additional examples of how semantic layers can help ChatGPT understand the meaning of language:

• Sarcasm: ChatGPT can identify sarcasm in language and generate a response that is appropriate to the context.
• Humour: ChatGPT can understand humour and generate responses that are funny and witty. As long as you know how to ask, if you say «tell me a joke», it's bound to be bad.
• Emotions: ChatGPT can identify emotions in language and generate responses that are empathetic and sympathetic.

ChatGPT can use the semantic layers to understand the context of the question and generate the most accurate answer.

8️⃣ The importance of training data:

Imagine an artist painting a picture.

The quality of the materials he uses, such as paint and brushes, is fundamental to the final result.

ChatGPT training data works in a similar way.

What is training data?

These are large amounts of text that are used to train ChatGPT. This data may include:

• Books
• Articles
• Conversations
• Code

Why are they important?

The quality and diversity of the training data are crucial for ChatGPT to be able to

• Generate accurate and relevant responses.
• Understand different types of language.
• Avoid bias in your answers.

If the training data is of low quality or not diverse, ChatGPT can generate answers.

• Incorrect.
• Irrelevant.
• Biased.

Example:

If ChatGPT is trained on a dataset that only contains text from men, it is more likely to generate responses that are biased towards men.

9️⃣ The Black Box: A Mystery to be Solved

I will give an example with electricity so that we can understand better.

• Externally we can observe the flow of electricity through cables, plugs, devices and some privileged ones have felt it in our own bodies when touching bare wires 🤭.
• Interior, i.e. what really happens, The precise mechanics of how electrons flow and generate energy remains a mystery to many.

The ChatGPT neural network works in a similar way.

What is the neural network?

It is the central part of ChatGPT that allows it to learn and generate responses. The neural network is made up of millions of interconnected units, called artificial neurons.

Why is it a black box?

It is difficult to understand how the neural network makes decisions because of its complexity. We cannot directly observe how artificial neurons process information and generate responses.

What uncertainties does it create?

The complexity of the neural network may generate some uncertainties about:

• Biases that may be present in the neural network.
• The way the neural network makes decisions.
• The possibility of the neural network generating incorrect or inappropriate responses.

How is this problem being addressed?

Researchers are working on different methods to:

• Explain neural network decisions.
• Reduce biases in the neural network.
• Improve the security and reliability of the neural network.

Here are some unanswered questions about the ChatGPT neural network:

• How can neural network biases be identified and removed?
• How can the decisions of the neural network be explained in a human-understandable way?
• How can it be ensured that the neural network does not generate incorrect or inappropriate responses?

As researchers continue to work on these questions, we can expect the ChatGPT neural network to become more transparent and reliable.

Now that you've peeked behind the curtain, I invite you not to just sit there with your mouth open in amazement at the wonders of ChatGPT, as if it were a children's party magic trick.

Get your hands in the dough and start experimenting with ChatGPT as if it were your own personal crazy lab.

Want to see how a little tweak in your question can turn the answer on its head? Curious about how you can make ChatGPT change its answers with different feelings, just by changing the way you ask it? Well, in this article you have plenty to learn and experiment with.

Don't just stand there waiting for inspiration to fall from the sky. The real magic of ChatGPT is not in the answers it gives you, but in how you learn to make the most of it. Every question is an opportunity to hone your craft, to become that PRo using LLMs like ChatGPT.

So what are you waiting for?

Without practice there is no learning.

Experiment, play, challenge ChatGPT and yourself. And remember, the limit is your knowledge and your imagination.