Foreword

In the ever-evolving landscape of technology, the rise of Generative AI (GenAI) has dramatically reshaped the way we approach learning and development. When my son Raffy joined the Mathematics faculty at the University of Indonesia, he found himself at the forefront of this new era. One day, he asked me what is truly worth learning in the age of GenAI. My answer was clear: transformer neural networks and large language models (LLMs). These technologies are not just the foundation of the most advanced AI systems we have today, but they also represent the future of how machines understand and generate human language.

Raffy’s next question was just as insightful: "What language is the best to learn for this?" My response was Rust, a language that, despite being relatively new to the field of AI compared to Python, offers unparalleled advantages. Rust's language features—such as memory safety, concurrency, and performance optimization—combined with its growing ecosystem of crates like tch-rs and Hugging Face’s candle, make it a powerful tool for building and deploying LLMs.

I acknowledged that Rust is not as mature as Python in terms of adoption within the AI and LLM communities. However, in the GenAI era, where code conversion between languages has become trivial, this gap in maturity is less of a barrier than it might seem. At RantAI, they are dedicated to exploring and exploiting the capabilities of Rust’s current crates to match and, in some cases, even surpass the Python ecosystem. Our focus is on making Rust a viable and attractive option for AI practitioners by harnessing the full potential of its unique features.

Learning about LLMs is, at its core, a journey into mathematical abstraction and the architectural intricacies of transformer networks. With the advent of GenAI, the implementation of these concepts has become increasingly straightforward. GenAI tools allow us to focus on understanding the underlying principles, while the actual coding and implementation are handled with ease through AI-assisted generation. This approach transforms the learning experience, making it more exciting and less daunting. No longer do students need to memorize complex concepts or syntax; they can instead concentrate on deep comprehension and let GenAI handle the rest, particularly with techniques like Retrieval-Augmented Generation (RAG), which brings relevant knowledge to the forefront as needed.

I hope that this book, LMVR will attract and inspire a new generation of talents, not only in computer science but also in mathematics and physics. These disciplines are fundamental to understanding and advancing AI, and I believe that with the help of GenAI, more students will find the journey of learning LLMs both accessible and exhilarating. This book is not just a guide to Rust and LLMs; it is an invitation to explore the future of AI with curiosity and confidence. I encourage students and lecturers alike to embrace this new era of learning, where the synergy between human understanding and AI capabilities can lead to unprecedented innovation and discovery.