Large Language Models (LLM) Implementation in the Enterprise
It’s noteworthy to note that Open AI has stated it’s thinking about opening an app store specifically for AI apps (link). Other significant players, such as Google, AWS, and Hugging Face, are likely to follow suit. While the goal is clear—to become the go-to platform for the adoption of large language models (LLM) and generative artificial intelligence (GenAI)—there is also a chance that a corporate product released on the platform could overshadow the platform itself.
There are still questions about whether the AI Apps Store will be as revolutionary for mobile devices as the Apple App Store was for the iPhone.
It also got me thinking as to what are the LLM deployment options today for an Enterprise. I can think of at least 4 deployment architectures:
- Black-box LLM APIs
- Enterprise Apps in LLM App Store
- Enterprise LLMOps — LLM fine-tuning
- Multi-agent LLM Orchestration
Some observations reg. the 4 deployment patterns:
1. Black-box LLM APIs
This is a typical ChatGPT [1] example where we have access to an LLM API/UI through a black box. Other Natural Language Processing (NLP) basic activities, such as Knowledge Retrieval, Summarization, Auto-Correct, Translation, and Natural Language Generation (NLG), can be considered using similar LLM APIs.
The main method of communication on this site is prompts, and everyone is still honing their prompt engineering techniques.
In order to increase the likelihood of receiving the “right” response, prompts relate to tailoring the user input and giving the LLM API the appropriate context and direction. In order to discover the “right” prompt to elicit the “best” answer, prompt engineers conduct systematic experiments and record their results, which has contributed to the emergence of prompt engineering as a professional specialty.
- Business Apps in the LLM App Store
Since they offer an enterprise-ready solution, Enterprise LLM Apps have the potential to grow into a multi-billion dollar market and speed up the adoption of LLM. However, the same prudence needs to be exercised.
