Using Microsoft Word in multiple minds? An experiment with two workspaces in AnythingLLM.
When writing my last post using Word and AnythingLLM, I encountered a situation where I needed to work with different workspaces and settings. For instance, AnythingLLM provides a Chat mode with two configuration options: Chat and Query. When testing the effectiveness of RAG, I switched to Query mode in AnythingLLM. This allowed the model to provide answers only if document context is found in its vector database. In that experiment, I prepared two documents as the context for RAG. The main difference between these modes is how they handle answers. The Chat model provides answers with the LLM’s general knowledge and document context that is found. To set up this experiment, I created a workspace called "workspace1" to manage the settings and upload the documents. To set up this experiment, I created a workspace called "workspace1" to manage the settings and upload the documents. By comparing the results obtained from both modes, it becomes clear that the RAG has proven effective.
Why do I need another workspace? The answer lies in my desire to polish my writing with different styles. In general usage of LLMs, I don't require the uploaded documents as context, nor do I need the Query mode to constrain the model's text generation. Instead, I can rely on the default Chat mode, which allows for free-form writing without unnecessary constraints. This flexibility is where multiple workspaces come in handy. Having separate workspaces with distinct settings enables me to switch between styles and approaches seamlessly. It's convenient to have multiple workspaces available, especially when using AnythingLLM, as it allows for easy switching between them. My configurations are shown as follows.
Workspace1 in Query mode
Workspace1: documents in vector database for RAQ
Workspace2 in Chat mode
Workspace2: no document & no RAG
In brief, leveraging AI-assisted writing tools can greatly enhance your writing experience by allowing you to seamlessly switch between different writing personas for various situations. With GPTLocalhost bridging Microsoft Word and AnythingLLM, this feature is particularly noteworthy as it enables you to tailor your workspaces in AnythingLLM to suit your specific needs, thereby significantly boosting your writing productivity. Such convenient functionality is often lacking in cloud-based LLM providers, which may struggle to accommodate the diverse situations in writing with a single mode.