دانلود کتاب نحوه کار مدل‌های زبانی بزرگ

The skeleton of this book began to come together in the late 2010s when we saw several significant advancements in the field of artificial intelligence (AI) that we knew could soon lead to a breakthrough. The convergence of new types of computer hardware, the availability of vast amounts of data, and the growth of neural networks were rapidly converging to a tipping point where it was now possible for machine learning algorithms to accurately capture nuances of language and meaning at a surprising level of fidelity. With the right combination of breakthroughs, we knew this would enable an entirely new class of applications. We conducted research, built prototypes, had conversations with our colleagues, clients, and families, and sought to tell the story of how these advancements could change the world and the underlying techniques that made that possible. Then, at the end of November 2022, OpenAI released ChatGPT, and suddenly, this potential became a reality. By putting this technology into the hands of the public, anyone could gain firsthand experience by interacting with a chatbot powered by a large language model (LLM). As with any new technology, there was a lot of speculation as to what could possibly allow ChatGPT to interact with great fidelity and produce such high-quality output. We saw that, based on interactions with ChatGPT, people often assumed that there was more behind the curtain than truly existed, sometimes believing that we were truly on the cusp of general AI that could do anything. We found that our conversations shifted to what could practically be achieved using applications of LLMs, managing expectations, characterizing risks, validating behaviors, and negotiating the path between what’s realistic and what’s not safe or responsible to attempt. Fast forward to 2025, and we’re now firmly ensconced in the era of generative and agentic AI. We have seen a massive proliferation of models, applications, and capabilities and an explosion in the types of data we can work with. Each major vendor has a technology offering that incorporates an LLM, whether they are chatbots to talk to or agents that review our writing, help us write computer programs, or generate images. Many of these are controversial, leading to new conversations about data use and causing us to rethink our assumptions about the relationship between technology and creativity. Regardless, there are core principles that enable these applications, and our goal with this book is to describe these in a way that’s accessible to readers from all walks of life. Whether you’re a CEO, a machine learning engineer, a casual coder, or just the average person seeking to use this technology, we hope you’ll find something useful in this book that explains the algorithms and techniques that make LLMs work. It is a collection of our experiences working in the field of natural language processing, machine learning, and algorithmic research, where we set out to share our knowledge in a manner that is accessible to nearly everyone. Along the way, we will dispel some of the mystery, explain the limitations, and explore the implications of this fascinating new technology. We hope you’ll join us on this voyage.