Large Language Models, From Magic to Foundations.

The explosive appearance of ChatGPT last year immediately gave me the feeling that "something was going to change", even without a precise idea of the true impact it might have.

This hunch prompted me to take two immediate actions: on the one hand, to seek training that would allow me to understand and contextualize change in both the business and technological fields, and on the other, to begin researching on my own.

In terms of research, I chose not to delve into technical issues such as tools, frameworks or methodologies, but rather to gain an overview of the fundamentals and focus on integrating the tool as an additional resource in both my professional and personal life.

From Magic

• In just over 3 months I assimilated the fundamentals of Java from my knowledge of Python (no human would have tolerated my resistance to modifying strongly established patterns).
• Internet searches for useful information on some topics began to be shorter and shorter, not only because the model having the answer (please, always double-check any provided info), but for sure helping to find the 'correct' terms to search over.
• New games shared with my daughter and wife, from 'dynamic' stories to craft surreal illustrations together, you might think it's still technology but I'm glad she's now 'writing stories' instead of passively consuming content.

Walking the path

A month ago, with more training and, mainly, experience in its use, I considered it the right time to 'understand the magic' behind ChatGPT and start to rationalize both the fundamentals of the tool and the 'techniques' to achieve the best results.

This interest motivated me to start publishing a series of posts that were perhaps somewhat scattered and not entirely clear about my 'research journey' ( you can found them in my feed, if curios) aiming to account for the progress made and also to force myself to reach conclusions.

I finally reached a 'base camp', a milestone that serves not as the end of my journey but as a safe starting point for the next phases of my exploration. It's a place I can venture out from with confidence and return to if the path ahead strays too far from my next goal.

To Foundations

In other words (and aligning with the title, which turned out very well, if I may say so), these are the fundamentals that conclude this first stage and upon which I propose to deepen and continue investigating, now in a more formalized manner:

• ChatGPT is a tool, simply a way to leverage the real technology that is revolutionizing everything around it: the LLMs, presenting itself as one of the most basic applications, a general-purpose 'chatbot'.
• LLMs are not 'intelligent' in the strict sense; they do not understand what they do nor possess qualities that make them 'more human' than other algorithms (although it may seem or we wish to believe it). We could describe them as 'databases of both theoretical and procedural knowledge, accessible through the use of natural language'.
• LLMs do not evolve, do not learn new concepts, nor gain experience autonomously. An LLM with expanded capabilities can only be obtained by replacing the existing one. Our most effective tool to leverage their potential is to provide the right context for each objective.
• LLMs are not universal solutions nor solve problems on their own; they constitute the foundation upon which to develop new tools that, in combination with other elements, will assist us in various contexts.

Thanks for read :)

I hope this reflection is of interest; not so much as a valid conclusion per se, but rather as an invitation not to believe in dogmas or wait for expert opinion, but to experiment and draw conclusions for oneself.

This has been my first article, and I hope not the last, but it represents a first milestone reached, a materialization that with curiosity and perseverance, one can draw conclusions with enough foundation to build upon them.

I will continue to share my concerns and conclusions. I do not know if they will interest anyone or not, but this exercise is proving to be enormously enriching both in terms of motivation and for my ability to synthesize and reflect.

See you soon!!

Disclaimer: I'm also improving/refreshing my English, after 3 years working in spanish projects and almost without use it, so if you find any error please corrections are welcome.