This review evaluates Guglielmo Iozzia's book "Domain-Specific Small Language Models," which advocates for a paradigm shift from generalist large language models to specialized, fine-tuned small language models (SLMs). The reviewer argues that SLMs offer superior control, visibility, and cost-efficiency for narrow tasks compared to the hype surrounding artificial general intelligence.
- The book promotes moving from renting to owning intelligence and from centralized to distributed systems.
- It provides a framework covering fine-tuning, quantization, RAG, graph databases, parameter optimization, multi-agent systems, and production deployment.
- The reviewer notes the content feels slightly dated regarding current frameworks but remains a useful starting point for building SLM systems.
- Criticisms include an underestimation of the effort required for fine-tuning, limited discussion on evaluation harnesses, and insufficient detail on regulatory compliance and structured decoding.
The book is recommended for AI practitioners and managers seeking to build domain-specific systems independently of current AI hype, offering both high-level management insights and practical technical guidance.