Post-project thinking: Enterprise Intelligent Assistants for SME

The rise of LLM has demonstrated the promising future of developing commercial EIAs in SME environments. Below are several items for consideration.

1. Transformer architecture is the core. 

Transformer has been an overwhelming neural architecture for implementing modern EIAs. The LLM is also implemented based on Transformer in a large-scale scenario. It is easy to transfer the current SMEs-based application to an LLM-based framework because they basically share the same architecture. There are also open-sourced LLM available that make it feasible to make this transition. 

2. Trustworthy collaboration is the future.

Most SMEs are unwilling to expose their data to LLM for data security and business protection. However, they need to plug into the LLM to utilise the powerful generative AI tool. The dilemma could be solved by leveraging the federated fine-tuning mechanism upon LLMs. Therefore, trustworthy collaboration will be a promising solution for future use scenarios of SMEs.

3. Self-evolving of EIAs.

Self-evolving is a key challenge to most EIAs systems due to the dynamics of balancing the customisation and generalisation in a complex system, such as LLM. The recent development Retrieval Augmented Generative technique is the solution to this issue. It simply accumulates information and data by organising it as a new database or document base, and then the RAG technique can complete the intelligent tasks by leveraging the LLM and recent documents in the database. There are many discussions about self-evolving by updating the LLM or parameters in the AI models. However, it is impractical or expensive to control in SMEs.