Large Language Models: The Hard Parts — The Book AI Practitioners Have Been Waiting For
Artificial Intelligence & Machine Learning News
For more than a decade, Rebellion Research has closely followed the work of Jonathan Regenstein and Thársis Souza. (See: Large Language Models in Finance With Super Quant Dr. Thársis Souza, Jonathan Regenstein on Machine Learning & A Practical Guide to Macroeconomic Data With R/Python)
In the increasingly crowded world of artificial intelligence, both researchers have consistently stood apart—not because they chase hype cycles, but because they focus on the difficult realities of building reliable, scalable AI systems.
Their newly released book, Large Language Models: The Hard Parts: Open Source AI Solutions for Common Pitfalls, arrives at exactly the right moment.
The AI industry currently suffers from a strange contradiction. Large language models have become astonishingly powerful, yet many organizations deploying them continue to struggle with reliability, evaluation, safety, hallucinations, governance, retrieval, and production readiness. While countless books explain what LLMs can do, very few address the uncomfortable question every executive, engineer, and data scientist eventually encounters:
Why is deploying these systems so difficult?
Regenstein and Souza tackle that question head-on.
Rather than celebrating the latest benchmark scores or model releases, the authors focus on the operational realities of AI deployment. They examine the problems that emerge after the demo succeeds and before an enterprise can trust an AI system with meaningful work.
This distinction is crucial.
The gap between experimenting with ChatGPT and building an AI application that an organization relies upon every day is enormous. It is filled with challenges involving evaluation frameworks, retrieval-augmented generation, context management, safety controls, alignment, testing methodologies, structured outputs, and monitoring. These are precisely the issues that separate successful AI initiatives from expensive failures.
What makes Large Language Models: The Hard Parts particularly valuable is its emphasis on reproducible solutions. The authors avoid vague theoretical discussions and instead provide practical Python implementations built around open-source technologies. Readers are not simply told what can go wrong; they are shown how to identify, measure, and mitigate those failures.
The book arrives during a critical phase in AI’s evolution. The first wave of enterprise AI adoption was largely driven by excitement and experimentation. The second wave, which is now underway, demands discipline. Organizations increasingly need systems that are explainable, measurable, auditable, and dependable.
This is where Regenstein and Souza excel.
Their focus on evaluation is especially timely. In traditional machine learning, performance metrics are relatively straightforward. Large language models introduce a fundamentally different challenge. Outputs are often nondeterministic, subjective, and context-dependent. Understanding how to evaluate these systems rigorously has become one of the most important—and misunderstood—problems in modern AI.
Equally important is the book’s treatment of retrieval-augmented generation, long-context systems, and alignment. As organizations seek to deploy domain-specific AI in finance, healthcare, legal services, education, and research, these challenges become central to success. A model that performs impressively in a demo environment may fail catastrophically when exposed to real-world data and operational constraints.
The authors provide readers with a practical roadmap for navigating these issues.
Perhaps most refreshing is the book’s balanced perspective. Regenstein and Souza are clearly optimistic about the transformative potential of artificial intelligence, yet they refuse to indulge in the exaggerated promises that often dominate AI discussions. Their approach reflects the mindset of experienced practitioners who understand both the extraordinary capabilities and the very real limitations of modern language models.
For data scientists, machine learning engineers, software developers, product leaders, researchers, and executives evaluating AI initiatives, this book offers something increasingly rare: clarity.
At Rebellion Research, we have long believed that the future of artificial intelligence belongs not merely to those who build the biggest models, but to those who can deploy them effectively, safely, and reliably. Large Language Models: The Hard Parts embodies that philosophy.
As AI moves from experimentation to infrastructure, books like this become essential reading.
Many people have learned how to use large language models.
Far fewer know how to make them work.
Jonathan Regenstein and Thársis Souza have written a guide for those who intend to do exactly that.
Amazon.com: Large Language Models: The Hard Parts: Open Source AI Solutions for Common Pitfalls eBook : Souza, Thársis T. P., Regenstein Jr., Jonathan K.: Kindle Store
Join us: Cornell Financial Engineering Manhattan 2026 Ai & Future of Finance Conference : September 11 2026
