This presentation explores the potential of LLMs & GenAI in infrastructure management, focusing on current capabilities, limitations, and future developments.

1. LLM Models I'll explain how LLMs work, how they are trained, and their probabilistic approach. I'll illustrate difference through Python & Terraform code generation comparisons

2. Issues with IaC Generation

3. Context Limitations: AI might produce suboptimal configurations due to missing context, leading to inconsistencies or dependency issues.

4. Security Risks: Models trained on public data may propagate vulnerabilities (e.g., open ports) and bad practices.

5. Synthesis vs. Generative AI Generative AI creates code/content, while synthesis AI analyzes and combines existing information, like logs, to identify issues. Understanding this distinction is crucial for effective use.

6. Future Potential - Context retrieval AI’s full potential will be realized when it integrates comprehensive environmental context, including configurations and service interdependencies. I will make the distinction between classical RAG (retrieval augmented generation) and graph-RAG to create such context, and their current limitations.