Unravelling the Mysteries of AI with SwissGPT.

Introduction

Have you ever found yourself wandering in the artificial intelligence (AI) jargon labyrinth, feeling a bit like Alice tumbling down the rabbit hole? 

Well, let’s embark on a journey to unravel one of these intricate concepts together – the workings of AI chatbots like ChatGPT, Bing or Bard. 

And our trusty guide on this adventure? A humble SLM.

Now, you might be wondering, what’s an SLM? 

Well, an SLM is a Small Language Model, a far cry from the LLMs (Large Language Models) that everyone’s buzzing about. These LLMs are trained with billions of inputs over weeks or months and come with a hefty price tag.

Our SLM, or as I like to call it, a VVSML (Very, Very Small Language Model), is a different beast altogether. It’s not as flashy or as versatile, but it does one thing exceptionally well. It answers “The One BIG Q”: “How Does ChatGPT (and every chatbot) Work?”.

Our SLM is the mastermind behind our unique SwissGPT. Why Swiss, you ask? Well, it’s trained on just one song – the classic tune “Do, Re, Mi” from “The Sound of Music”, a film set in the picturesque landscapes of Switzerland, hence SwissGPT. It might seem quirky, but trust me; it serves its purpose beautifully.

So, lace up your metaphorical hiking boots, and let’s embark on this enlightening journey.

A Lunchtime Epiphany

The other day, over a casual lunch with my daughter, our conversation meandered into the realm of AI. 

Yes, I confess, my passion for AI tends to sneak into everyday chats. 

As we exchanged bites and ideas, I found myself painting a picture of how AI chatbots function. 

And you know what? She grasped it in a heartbeat. 

Now, she’s a bright spark, but perhaps the way I explained it played a small part too.

The Simple Magic of SwissGPT

Let’s kick off with a playful thought. 

Imagine SwissGPT as a diligent student who has only ever studied one thing – the song “Do, Re, Mi”. 

If you were to prompt SwissGPT, “Do, re, mi”, it will eagerly respond with “fa, sol, la”. 

Why, you ask?

Because it’s learned the pattern from the song and is doing its best to predict what are the most probable words to come next based on what it has learned.

Quite straightforward, isn’t it?

SwissGPT’s Grand Tour of Songs

Let’s stretch our imaginations a tad further. Picture SwissGPT embarking on a global music tour, soaking up every song, in every language, sung by anyone and everyone. It’s like having the world’s most eclectic Spotify playlist. 

But what happens when someone, let’s say, forgets the lyrics and sings “la, la, la” instead? 

Well, if you were to prompt SwissGPT with “When I find myself in times of trouble”, it might respond with “la, la, la, la la la la”. Why? Because it learned from a Beatles fan who didn’t quite remember the lyrics to “Let It Be”.

SwissGPT’s Echo in the Alps

Here’s a little nugget to remember – SwissGPT doesn’t understand the difference between the “la” that refers to a note in the “Do, Re, Mi” song and the “la” that’s just some random text replacing the one in “Let It Be”. 

This is because Swiss GPT, like all GPT models, doesn’t understand words or text. It just convert words or parts of words (tokens) to numbers, does its magic on the numbers and knows how to convert its numbers back to humanly understandable (hopefully) text.

In our simplified example, “fa sol la” would be transformed into [6904, 14017, 2474], and “la la la” into [2474, 2474, 2474]. The note “la” [2474] is represented identically to the “la” [2474] in “la la la”, which serves as filler words when the original lyrics are forgotten.

Our SwissGPT has no clue about the different meanings of the two, it’s just numbers.

It’s a bit like the echo in the Swiss Alps – when you shout “Hello”, it answers “ow, ow, ow”. But remember, that’s not the mountain answering; it’s just the echo of your voice.

In our SwissGPT, the echo of all the voices it learned from.

SwissGPT and Ethics

Let’s spare a thought for this hypothetical SwissGPT without stretching the metaphor too far. If you feed it bad words, and the echo goes “..it”, please don’t blame the mountains. 

Ethics isn’t about AI, SwissGPT or ChatGPT, Bing or Bard; it’s about what we put into them and how we use them. 

The entire discussion about ethical AI is just an exacerbated discussion about Human ethics, and while that discussion is paramountly essential, don’t blame the AI if we transgress our human limits with it; just let us look at the one in the mirror.

For the More ‘Serious’ or at Least ‘Eager to Know More’ Readers

If your curiosity has been piqued and you’re itching to delve deeper into the world of AI, I’ve got a couple of extra layers of explanation that will make you sound like the life of the party at your next social gathering (or at least the most informed).

There’s the ever-charming Andrew Ng and his courses on Coursera. Here’s the link to his free introduction about AI, “AI for everyone“. It’s like a friendly guide through the dense forest of AI knowledge.

Then there’s a YouTube video of a particular kind, in which Kyle Hill explains LLMs in less than 30 minutes. It’s all in there and delivered in a very particular style (try to look beyond this) because it gets you to a proper understanding of how LLMs and bots like ChatGPT work.

If you’re itching for more, the world of AI is a treasure trove of knowledge waiting to be explored. But it does make you wonder, do we really need to know every thread and screw of the nuclear plant to flip a switch and bask in the light it provides?

AI and the entire ecosystem are here, and they are here to stay. Let’s learn, harness them and build a brave new world for the better of all.

Conclusion

Demystifying AI doesn’t have to be a Herculean task. 

By breaking down complex concepts into simpler ideas, we can all better understand these fascinating technologies. 

So, next time you interact with SwissGPT or any more serious AI, remember, it’s like an echo in the mountains – it repeats what it has heard but doesn’t understand the meaning behind the words. 

We are still far away from sentient AI’s. I’m not saying there will never be one, I’m just saying that it’s nice and reassuring to understand how AI works today, even at this elementary level.

The danger does not come from AI itself. It will come from a combination of 2 factors: how some will use it and the ignorance of the ones on whom it is used.

Education & Learning, will be the cornerstones for a safe environment for the next generations. Education is what we can give, but Learning will only come with the right motivation and inspiration.

That’s the mission of AI Leadership, that’s our mission.

Yours in AI

Luc

TL;DR

Welcome to the world of AI, where we unravel the mysteries of chatbots like ChatGPT using our trusty guide, SwissGPT. This Small Language Model (SLM) is trained on just one song, “Do, Re, Mi” from “The Sound of Music”. It’s a simple but effective way to understand how AI learns patterns and predicts what comes next.

Remember, AI models like SwissGPT don’t understand words or text. They’re like an echo in the mountains, repeating what they’ve learned. When it comes to AI ethics, it’s not about the AI, but about us – what we feed into them and how we use them.

Curious to learn more? There’s a wealth of knowledge out there waiting to be explored. But remember, the proof of the AI is in the prompting. Happy exploring!