What are LLMs & How Are They Changing Our World?
And everything you need to know about how they work.
Nowadays, everyone and their pet has started using LLMs, but what is going on behind this technology? Chatbots, Virtual assistants, and AI-powered tools, LLMs are already everywhere, and they’re just getting started. LLMs are shaping the world we live in. You might be wondering what LLM is in the first place, right?
Large language models (LLMs) are AI-based systems that are trained on a vast amount of human-readable data such as books, and articles which enables them to understand human language and respond just like humans do by getting familiar with their communication patterns. Different AI models such as Chat GPT, DeepSeek, Claude, and many others are built using this technology called LLMs.
The world has become so heavily dependent on AI models that it almost seems impossible to imagine a world without this luxury. Whether you need someone to help you debug your code, automate your emails, or teach you whatever you want, AI models can literally do everything.
Today we are going to learn about what LLMs are, their positive and negative impacts on society, and what the future of LLMs might hold. Anyway enough talking, let's get to the juicy part folks!
What are LLMs
Imagine you have a movie script that contains the dialogues of 2 people. Let's suppose you are the 1/2 people and your dialogues mostly include questions that must be answered by the other person but here is the twist. The dialogues of the other person are torn apart.
Now, imagine there is a super machine that can take any script containing text and can provide a sensible prediction of what will come next if you give this machine enough input. In a long enough time frame, it can generate an output that will almost be perfect.
This is exactly how LLMs are working underneath the hood. They are trained on immense amounts of data enabling them to understand and generate human language, like English.
As a matter of fact, whenever we interact with any chatbot, it is essentially predicting what to say based on the training material it had and our input.
Furthermore, you can define LLMs as:
A sophisticated mathematical function that predicts what word comes next for any piece of text.
Types of LLMs
We have understood what an LLM is but the show doesn’t stop there, there is much more to the world of LLMs including different types of it, and each type is unique in its own way. You must be pretty curious at this point to learn about them right? Let us get to the main part without wasting any more time.
Autoregressive models
Imagine you’re building a sentence, one word at a time. You already have a few words in place, and in order to further continue the sentence, you need to predict what word should come next based on what you’ve already written.
This is essentially how Autoregressive Models work. They are designed to predict the next word based on the context of the words that came before it. You can kind of think of these models as Dr.Strange of the real world.
Models such as GPT-3 fall into this category. Although they are really useful, they come at a cost. Apart from that, there is a downside in this particular model, which includes irrelevant responses and repeated phrases. You might have also experienced it sometimes when using the Open AI’s GPT-3 model.
Masked Language Models
These models are a bit different from autoregressive models. Think of them as a model that tries to fill in the blanks in a sentence but with a twist. Instead of predicting one word after another (like GPT). The job of this model is to predict the missing words, based on the text not only before those missing words but also after the missing text.
It looks at both the words before and after the blank that it is trying to figure out. Unlike the autoregressive model, the masked model doesn’t just look at the words before, it looks at the entire context (before and after).
BERT is a famous example of this type of language model. I can give you a small example to further clear any misunderstanding you have.
For example:
Input: The cat sat on the __
Predicted output: mat (based on the entire context).
Encoder-Decoder Models
These models are a specific type of Transformer-based architecture (More on this term later) that are designed to handle tasks that involve translation, summarization, or question answering, where there’s a need to both understand and generate language, one after the other.
An Encoder-Decoder model consists of two main components:
Encoder: This part reads and processes the input that involves a sentence or phrase. Its job is to compress this information into something called contextual embedding. In simple words, it means that it is responsible for grasping the main idea of a sentence.
Decoder: After the encoder has completed its part, the decoder steps in and uses that information to generate the output. Its job is to predict one word at a time until the full response is complete, like slowly answering a question by using the information it has
in its mind.
AI Models like T5 and BART fall under the category of such models.
Pre-trained and Fine-tuned Models
Pre-training is the first step where a model is trained by giving it a vast amount of data in the form of books, articles .etc to learn different language patterns, grammar, and general knowledge about the world. This step is very essential, it's like we are molding it to become as close to a human being and this is our first step.
Our main goal is to teach the model to understand and generate human language. You can think of it as teaching a student the basics of a subject before asking them to do a specific task.
Fine-tuning is the second step that takes place after pre-training. It's the process where a pre-trained model is further trained by giving it data related to a specific task so it can become better at it. This allows the model to specialize in a particular niche. It can be different things such as answering questions, summarizing text, or translating languages.
Our main goal is to train the model into a master of one specific task instead of being a jack of all by exposing it to examples related to that particular task. It's like training a student on a specialized topic after he or she has mastered the basics.
Multilingual Models
Multilingual models are different from all the models so far. Their job is to understand and generate text in multiple languages which is mind-boggling, isn't it? These models are trained on a wide variety of data containing text from various languages. This ultimately allows them to handle tasks in several different languages at the same time.
Our main goal by creating such a model is to enable a single model to work across different languages. By making such a model there is no need for separate models for each language. You can think of such a model as that brilliant student who learns multiple languages to communicate with people from different countries (I bet you have such a nerd classmate who can speak multiple languages lol).
Architectures Behind AI Models
You can imagine AI as a 2 faced dragon because AI isn’t limited to the world of text. It has two major sides one that understands and generates text (like ChatGPT) and another that creates images (like Stable Diffusion | Midway journey).
Text-based AI uses a Transformer model and its job is to predict and generate words. Meanwhile, image-based AI uses Diffusion Models, and its job is to convert random noise into detailed images.
To really understand how these work, we will take a look behind the curtains of these AI models. So without wasting any of our time, let us get to the interesting part.
Transformer Models: The Backbone of Text-Generating AI
Transformer models are the heart of AI models, this architecture is the reason why AI models are able to understand natural language and in return generate human-readable and understandable text.
To understand the workings of a transformer in a really simplistic way, I am going to use an analogy here about books. Imagine you're reading a book. Instead of reading it word by word in a sequence, you quickly scan the entire page, which allows you to grasp the main details all at once. That is exactly how Transformer models work in a nutshell. Instead of processing words one by one, they look at the whole sentence or even the whole paragraph at once to understand the context better.
The transformer model’s working might sound similar to you because of our discussion about different types of LLMs. The reason for that is that different AI models are mostly based on the Transformer architecture and they all have different qualities of the Transformer. Each AI model then uses the quality it inherited to become a master in that specific quality.
Autoregressive Models (GPT), Masked Language Models (BERT), and Encoder-Decoder Models (T5, BART) are all based on transformer architecture. You can think of the architecture itself as the father while these different AI models as its children, each child inheriting a specific quality from the father.
Diffusion Models: The Backbone of AI Image Generators
Apart from the world that involves the text-generating LLMs. There also exists a world in which AI models are capable of creating realistic images, art, and even videos. All this is possible due to the architecture known as the Diffusion Model.
To understand the workings of a diffusion model, I am going to use an analogy involving an artist assigned to complete a painting. But here is the twist, instead of working on a painting from scratch, he works on a painting containing different random splatters of colors.
The artist slowly removes the noise, bit by bit, until a clear image emerges from the dark. That’s essentially how diffusion models function. They start with random noise and gradually improve it until it transforms into a recognizable image.
AI models like Stable Diffusion, MidJourney, and DALL·E are built on the diffusion model architecture. Just like the inheritance takes place in the case of transformer architecture, the same thing happens here as well. Just like how transformers turn patterns of text into meaningful human-readable sentences. In the same way, diffusion models are used to convert random noise into meaningful images.
Real-Life Applications of LLMs
LLMs are playing a major role in shaping the world we live in, it is changing how we use technology. Whether it is a matter of customer support, automating tasks, or even scaling a business, LLMs are everywhere.
Tasks that took hours or days to accomplish, can be completed in a matter of seconds with the help of AI. LLMs have revolutionized the world similar to how the Internet brought a change in the world when it was introduced to the average Bashir of the world.
You must be wondering that, it seems like a great technology but where and how is it used right? Let me shed some light on it.
Content Creation
The capabilities of AI are limitless, understanding and generating natural language text is just one of its abilities. LLMs can do much more than just write, different tools allow users to create stories, generate articles, and summarize information instantly.
Nowadays, bloggers have AI co-writers who never get tired and sleep. So when their main writer is on break and they don’t want the content to stop, with the help of their co-writer they can publish as much content as they want which is a game changer.
Although it is easy to identify AI-written material, as time progresses, it is getting much harder to do that as well. AI is improving rapidly making the line between human-written and AI-generated content blurry.
AI Influencers
AI influencers, also known as virtual influencers, are computer-generated characters.
Such AI-generated personas are a result of the combination of technologies like AI models and computer-generated imagery also known as (CGI). Most of these virtual characters have become a part of reality due to the work of two leading AI-generating tools, Midjourney and Stable Diffusion.
These tools usually implement text-to-image generation. This means the creator enters a text prompt, and tools like Midjourney output a result for that given prompt. The results are then further refined according to the creator’s liking. Once a virtual persona has been generated, the creator sets up a social media account and customizes the AI character according to their target audience.
The famous Instagram AI influencer called Aitana Lopez is just one example of this. People are using such a trend to their benefit which also involves making tons of money in real life. Fascinating isn’t it?
LLMs in Cybersecurity
LLMs are changing the online security game with each passing day, enhancing both defensive and offensive strategies. Instead of just analyzing threats the old-fashioned way, LLMs can now predict, prevent, and automate security tasks in ways that weren’t even possible to imagine before. Let me give you a bit of insight into some tools that are utilizing AI to play a major role in the space of security.
Snyk: This tool helps developers automate the detection of any kind of weakness in their code. With LLM integration, it not only catches the issues but explains them in plain English and suggests fixes.
In simple words, it acts like a big brother (Snyk) guiding his small brother (user) in fighting the bad guys (hackers) at every step of his life by fixing his mistakes (weakness in the code).
X-Bow (Upcoming): Imagine you’re playing a video game where waves of enemies (hackers) keep attacking your fortress (system). Normally, you’d have to manually build defenses, repair broken walls, and strategize every move.
X-Bow is like an auto-defense system. Instead of waiting for you to react, it will automatically do everything. If X-Bow delivers on its promise, it won’t just help companies fight off cyber threats, it will surely be a weapon of great value in the space of security.
Chatbots & Virtual Assistants
Imagine having a butler with you 24/7 who doesn’t sleep, eat, rest. It only follows your commands and never does anything that can make you angry. That is quite literally the definition of chatbots and virtual assistants in the modern era.
Think of them as digital genies, trained to understand what you need and then grant you your wishes. Whether you want it to provide Customer service, guide you on how to play video games or behave like a friend who recommends the best movies to watch and food to eat according to your own taste, it can do everything. It can even act like your teacher and teach you concepts in which you are facing difficulty. It can quite literally do everything which is so fascinating.
They’re like friendly, tireless workers who are always listening, always responding, and always making life a little easier.
Further Applications of LLMs
The applications about which we discussed earlier are just a few examples of how LLMs are being used today. Honestly, we could talk about them all day. From healthcare to entertainment, education to finance, LLMs are making their way into every niche.
Whether it's helping doctors analyze patient data, guiding students through complex problems, or even generating creative scripts for movies, the possibilities are endless.
Nowadays, the applications of LLMs are growing at such a fast rate. LLMs are being utilized in different jobs like AI Personal Stylists, Coding Sidekicks, and even AI Lawyers. The application of LLMs is endless, that is how heavily dependent the world is on it.
How LLMs Are Changing Our World
Before AI became a part of our lives, things were different. Everyday tasks that we now take for granted such as getting quick answers, a guide on how to build something, or translating languages. All of it required tons of human time and effort. Don’t get me wrong, technology was growing, but as soon as technology like AI got introduced to the public. It changed everything.
Today, LLMs are everywhere. They help us with everything from answering questions to making decisions. We interact with them daily without even realizing it. They have become a part of our life.
There were times when developers used to visit websites like Stackoverflow to find solutions for the problems in their code but that era is gone now. People now open their favorite chatbot in their browser, tell it the problem they are facing and boom AI just solves it in a second. Banging your head in the wall for 4 days just to realize that a semi-colon is missing on line 48 is not a problem anymore.
But with all the good AI brings, it’s also changing things in ways we didn’t expect. As LLMs keep improving, they’re having a much bigger impact on our society. They are influencing the way in which we work, think, and even what we value and this influence isn’t always a good thing.
The big question is: what are we gaining, and what might we be losing?
I’m not saying that AI is turning our world upside down in negative ways. Of course, there’s a bright side as well. But the truth is, this is a complex and controversial topic. In order to understand the full picture, we need to take a step back, breathe in, and look at both the positive and negative impacts of AI.
Positive Impacts
Let's forget about all the negative impacts of AI so far and simply just focus on the actual good things it has brought into our lives. If we actually do that, it becomes hard to count the blessings of AI, it simply is a true example of disruptive technology which had a very big impact on the world. What kind of impact? Let’s talk about that.
Revolutionizing Healthcare
The human body is like a puzzle, filled with silent killer diseases inside it, which are near impossible to detect before something serious happens. For years doctors relied on experience, guesswork, and outdated systems to find them. But guess what? The game has completely changed since AI stepped in.
Take breast cancer, doctors fail to spot early signs of breast cancer in 20/100 of cases. This is not because they’re careless or lack skills, but because human eyes aren’t built to catch every microscopic detail.
AI has simply changed everything in the healthcare industry. It sees what we can’t. AI can scan medical images and detect anything out of the norm in the human body years before symptoms appear.
Cancer is just one of the many examples. AI-driven diagnostics are predicting heart attacks before they happen, analyzing retinal (scanning a part of the eye) scans to detect diabetes. Furthermore, it can even read brain scans to detect early signs of a disease called Alzheimer.
AI is like the infinity stones in the Doctor’s hand instead of Thanos’s hand. It’s giving them superpowers. And in the war against disease, that’s the closest thing to a cheat code we have ever had.
Improvements in Education Sector
For decades, education has been stuck in the same loop that includes 1 unpassionate
teacher who is just there for the salary and a dull classroom filled with bored students.
Along with that, the system is designed to make everyone work in the same way and not everyone is designed for it which leads to different problems. Due to this system, some students just quit, while others fall behind and lastly, there are the students who struggle and because of it, they get treated like an outsider.
AI is changing that. And I definitely am not talking about things like automating grading or recommending YouTube tutorials. I am talking about building a system that actually understands how each student learns.
AI is transforming education by making learning personalized. Different Platforms like Khanmigo and Duolingo make sure that each student gets the support they need without falling behind in the line. We also have AI tutors who provide 24/7 guidance, converting complex topics to easier topics for those who can’t afford private help.
You thought that was it? Sigh! We also have different AI-powered tools to help those with disabilities by offering speech-to-text, text-to-speech interfaces, and much more. The old system that tried to fit students into the same box regardless of their shape is being torn apart, thanks to AI.
Change in Climate
Our old planet called Earth is heating up, ice caps are melting, and extreme weather is becoming the new normal. The Governments move too slow. The big Corporations are busy worrying about making money. But AI doesn’t need politics or profit. It simply runs the numbers and boom! Everything is all in its control.
The different AI models that are used for weather forecasting, are probably used by everyone occasionally. You want to go on a vacation and want to know whether it's going to rain? One search and it's all set. AI models can predict quite literally everything except the future including hurricanes, earthquakes, and floods before they happen.
For years, tackling climate change felt like an impossible fight. But thanks to our boy AI who neither gets distracted nor gets tired. It simply just keeps working like a menace. Running simulations and analyzing the weather, AI is literally locked in.
Improvements in Cybersecurity and Fraud Prevention
Before AI entered the game, the world of cybersecurity and fraud prevention was very different. It was like trying to catch a bullet with your bare hands. The world was filled with slow computers that didn’t respond until something serious had happened. It was like these machines were asleep and cybercriminals knew this due to which they took full advantage.
The world of online security was a disaster and fraud prevention was no better. We relied on old-school rules like flagging transactions when they exceeded a certain amount or if something was happening from an unusual location. The governments thought they were smart but guess what? fraudsters were always two steps ahead. They always had a better scheme which allowed them to slip right past those basic so-called rules.
AI’s impact on cybersecurity and fraud prevention is nothing but a true definition of revolution. Different AI tools have the ability to spot unusual behavior and not only that. It can also predict cyberattacks before they happen. It has the ability to literally catch every kind of threat which allows it to secure whatever it is securing from hackers.
On top of that, if a hacker manages to break into a system somehow, AI steps in right away. It locks things down and fixes the issue without waiting for people to act.
Similarly in the field of fraud prevention, it is nothing less. AI tracks real-time spending habits, flagging suspicious activity as soon as it happens. Imagine if someone tried to steal your credit card info while you were on vacation and tried to spend all the money, you would get in trouble, right? But guess what AI’s got your back.
Negative Impacts
Well, I have sprinkled enough glitter into your eyes about all the positive impacts of AI. You probably thought that there aren’t any bad impacts of AI on society right? Surprise! There are tons of dark sides to this technology. I am not saying AI itself is bad or that some hidden terminator side of AI exists. Just saying that everything always has a good and bad impact on us humans. Anyway, let me give you some spicy insights into the impacts of our boy AI, so you can really see what I’m getting at.
Discouragement of Problem Solving
To fully grasp my point, imagine you have a time machine and you use it to travel back to the era when AI did not exist yet. In order for humans to solve a problem, they had to rely on their brains. They had to use skills such as critical thinking. They had to apply effort and hard work to each task. But everything has changed now, people have literally stopped using their own brains to solve a problem.
For example, we have Bashir who faces a problem. Instead of trying to solve it, he throws his problem to the chatbot and asks it to spoonfeed him the solution, and boom! The problem is solved. Bashir feels happy. But what Bashir doesn’t know is that he is damaging himself by relying too much on AI.
Similarly, for instance, let’s take a look at the current state of Software developers. The so-called ‘Tech Wizards’ of this era who are responsible for basically building quite literally everything in the software realm. Their coding skills are declining because of excessive use and dependency on AI. Is AI to be blamed for this? Of course not! We ourselves are supposed to be blamed for this.
Our brain is basically a muscle, the more we rely on AI, the weaker our brain will get which will ultimately affect our own thinking. Instead of training our brains, we’re letting machines do all the thinking for us. The more we lean towards AI doing the work for us the less we exercise our intellectual muscles.
You can think of AI as a very sharp knife, we can use it to cut vegetables and make a delicious dish or we can use it to cut ourselves. At the end of the day, whatever we use it for either good or bad. That is all on us.
Job Displacement
Yes, you heard it right. AI is coming for your jobs. It's happening faster than anyone predicted. Like, come on we’re literally talking about machines that don’t need coffee breaks, sleep, family time, vacations. It can work 24/7 and no human can compete with it.
Many workers have already been kicked out of work in different domains such as Customer services, and factories. And this is not it, there are literal self-driving trucks that are replacing drivers. Think about it, a self-driving truck needing no sleep or breaks being used for delivery? That is quite efficient but it ain’t good for an average person.
Quite recently Mark Zuckerberg hinted that Meta could begin replacing software engineers with AI as early as 2025. Similarly, Nvidia CEO Jensen Huang feels that AI will replace programmers and no one will require C or Java to do coding. Should the developers who put countless hours of effort into programming just quit? The real question is what exactly is the truth?
Although, tons of employees working in different job roles are indeed at risk of being replaced by AI. But this is not entirely true in the sense that if you are one of those people who are adapting to the modern era by using AI in the right way instead of being used by it. Then you shouldn’t worry about any of this, because you, my friend, are in the safe zone.
So a general advice for all of you whether you are a programmer or not. Maybe your current job involves some other entry role. Regardless of your current job, you need to start upskilling yourself because that is the only thing that can allow you to secure your current self and your future self.
The Deepfake Dilemma
Welcome to the modern era, an era where we cannot trust what we see, everything that you see on social media might just be an illusion. AI is turning reality itself into an illusion. Everyone has heard the phrase "seeing is believing" right? Well, AI just smashed the phrase into two pieces. Much thanks to the deepfake technology.
You might have heard the term deepfake many times but what is it? Deepfake is a technology that uses AI to create ultra-realistic fake videos, images, or audio which can literally fool you into believing something truly false. Nowadays, we cannot even trust things that we see on the screen anymore. With deepfakes, reality becomes a question mark.
I myself have seen so many fake-generated memes involving Donald Trump and some random alien shaking hands and all kinds of stuff that makes you laugh so much. I get it, it's funny to use technology in different ways, but not everyone uses such technology to make people laugh.
There are real dark people out in the world with ill intentions. Intentions that do not involve making people laugh which can be really dangerous when you think about it. Like, imagine that you can clone anybody’s face, or voice and do whatever you want with it, dangerous, right?
AI gave us great power to create incredibly realistic content. But I think we forgot the responsibility part (lol).
Apart from jokes, it truly is our responsibility to use such power in the right way. It’s our job to ensure we don’t warp reality to the point where we can’t even trust our own eyes anymore.
Loss of human control over decision-making
We humans are a prideful species. Taking pride in everything we do. We love to take pride in thinking that we are masters of our destiny, controllers of our fate, and sailors of our ship.
But the real question is, are we in control anymore? Because the way I see it, is that our big fella AI is the actual controller and we handed the control to it ourselves. We are the ones who put AI in charge.
Every single day, more and more decisions are being handed over to different computer algorithms. Whether it is what we see on our Instagram feed or who gets hired by the firm, every single decision is made by AI. We are not the ones making the big calls anymore. All of it is done by AI.
The scariest part? AI doesn’t think like humans. It doesn’t have emotions, empathy, or an understanding of the greater picture. It simply just processes data and we are allowing AI to make decisions for us. When we do that we often find ourselves trapped in a system we can’t understand or control.
The more AI evolves, the more riskier the situation becomes. Big decisions are being handled by some technology which greatly impacts our lives and society and it is something that should be stopped. AI may solve problems, but it also leaves us with a fundamental question:
Are we letting technology steer the ship, while we’re just along for the ride?
How LLMs Might Shape Tomorrow
Large Language Models have already shaped reality and rewritten the way in which we interact with all kinds of technology. Whether it includes tasks like automating customer
service or assisting in different scientific research, AI is found everywhere. But the real question that we should be asking ourselves isn’t just what’s next? It’s who will be in control?
On the one hand, LLMs could really be handy for us in the future by making the knowledge easier to access and understand for the average Bashir, personalizing education, and helping
to teach it to people who want to learn. Not only that, it could also help businesses run more smoothly by automating different tasks.
Like actually imagine AI-powered legal assistants making laws easier to understand. The truth is we can talk about the different benefits that AI can provide to us in the future all day. But with all these good things, the bad things also should be acknowledged.
Here is the catch, the more we rely on LLMs, the less we think for ourselves. When AI itself can write, and even think for us. Then where does the part go where our role is involved in all this? Where do the qualities such as human creativity, problem-solving, and critical thinking go?
And then lastly we have the most important factor. Who truly controls all these models? Right now, a handful of corporations and governments are pulling the strings for all these. But if left unchecked, LLMs could reshape reality itself, deciding what information should be given more priority and what should be fabricated and what not.
Will LLMs stay as tools that we control or will they quietly shape our world without us noticing? Truth is all of it depends on how we will use, and regulate such technology in the future.
Summing it up
Today was a special day. We dove underneath the hood of this mysterious technology called LLM and learned how it is being utilized in the world we live in. More importantly, we discussed the different ways it impacted society, of course in both good and bad ways.
From personalizing education to completely shifting the medical realm, it greatly impacted society in a good way in almost every niche. But of course, due to it performing tasks with such high efficiency, it also poses a risk to different job roles. But again as I mentioned previously, if we actually put in effort, stop being lazy, and upskill ourselves then we will not need to worry about AI replacing us.
Rather than AI replacing us, we can use this technology to help us in our work and perform even better. At the end of the day, it all comes down to the person using AI in the right way.
If you manage to do that, then you can quite literally become an unstoppable force.
Lastly, I would like you to recall that although we did cover many good and bad impacts of LLM on our society, the truth is that all the information you read is like a drop of water in an ocean. This is where the part of your own research kicks in, so I highly encourage you to go out and uncover the truth for yourself.
If any of you found any value in reading this, then I highly recommend you share feedback over through email at admin@sknexus.com. And don’t forget to stay safe and stay secure.
Further Learning
 | A guest post by
|