Understanding the AI-Powered Guide to Murshid Sam's Writings
Artificial Intelligence (AI) is a source of both curiosity and concern. This document looks to share some background, and the benefits and challenges of AI as a guide.
Intro to Artificial Intelligence and Large Language Models
In 2017 a breakthrough in neural networks opened the door to a wave of progress, leading to Large Language Models (LLMs). This technology reverse engineers how language maps to meaning, and can provide an almost human-like prediction of what words are most likely to answer any given question. The popular AI services (OpenAI, Anthropic, Gemini, etc.) are trained on vast amounts of the world's information. There are many reasons people are concerned and fearful of this technology, including the illegal use of intellectual property to train the models, the potential for bias and control of information, and the loss of work resulting from the use of AI. Many also argue that AI is simply a tool being wielded by humans, and can be used beneficially.
How the "AI Guide" works
This guide, in the form of a chat interface, is meant to help explore meaning in this large archive of Murshid Sam's written words. The entirety of Murshid Sam's works has been uploaded into a private "Vector Store" (a way to extend the knowledge of the AI service), which provides deep insights into the writing. In addition to the added writing, extensive training data is added to instruct the intended behavior (e.g., instructions to only answer questions about the archive and not stray into general information). The training instructions are part directives (intentions to behave in a certain way) and part reactive rule-setting (based on observed behavior we would like to encourage or change). This is where sharing negative experiences can be helpful as we can tune the rules to address issues.
We've incorporated "citations" into the AI Guide tool, which means most responses will try to reference the most likely source(s) of the information. While this works reasonable well, it's often difficult to find the exact sources as AI does not pull content from the documents directly.
The chat maintains an understanding of the conversational context. For example, if the AI answers a question, you could reply "why," and it would respond appropriately. Also be aware that each conversation is cleared when you navigate away from the chat and come back.
The goals of the guide
For anyone new to Murshid Sam's work, it can be overwhelming in both its volume and density. Depending on where you begin, there is no clear introduction to many esoteric concepts, historic or cultural references, or even names of people, places and experiences. One of the benefits of AI is that it can see the holistic picture of all the writing as a web of connections.
For more experienced followers of Murshid Sam, the tool may help connect ideas in new and unexpected ways. It could also be a tool for helping explore ways to translate esoteric meaning into accessible language for the uninitiated.
How to get the best results
The best way to use the AI Guide is to highlight a word or short phrase from Murshid Sam's writing. A window will appear with the option "Discuss." This will open the AI Guide with both the selected text and the page context, with the goal of the AI providing a clear overview of the meaning and context of the words.
In general, the best results are asking to explain terms or ideas that are not easily understood or require more context.
Asking for additional detail or finding related concepts is also very useful.
Try not to ask for specific names, quotes, or events, as the results can vary (see challenges below).
The challenges
LLMs have fundamental flaw when it comes to the representation of written information — the words chosen are not pulled from the source content, but based on the probability of the words being correct. This is most notable when requesting a quote or specific name of something or someone. In general, the more data surrounding an idea, the more likely it will be correct. In the case of very obscure information, specific inquiries are likely to be wrong to some degree. LLMs have no way to self-reflect on the quality or accuracy of the information. In addition, LLMs are highly authoritative — meaning they will always sound 100% certain, which can lead to a lack of trust.