Yet, leveraging this information effectively remains a challenge. A recent project brought the limitations into sharp focus when working around the chat token limits (as well as managing budgets and resources). It reminded me of our own cognitive limits (which are many), and gave inspiration for an innovative solution: a new type of database to pair with Large Language Models (LLMs), hinting at a more organic database concept that follows several patterns of our information retrieval.
In early experiments, we had marked success with embeddings, but overall found the costs prohibitive to scaling into the solutions we want to offer within our products and client solutions. This drove some creative work, in part (super)powered by our idea accelerator tool – Spryntify.
Key points include:
- Designed specifically to work within token-limited environments
- Optimised – stores specific knowledge only, removing anything generally available
- Self-evolving
- Conflict resolution strategies
- Drastically reduced costs vs. embeddings
The existing ‘trigger’ project used ChatGPT and MySQL, among other tools, but this innovation aims to significantly elevate the game while making developers’ lives easier, and embracing the concept would remove all limits on scaling.
What do we mean by Organic Database?
Largely, a more natural and imperfect and verbose information retrieval. Data is split between transient and absolute, to distinguish between fluid knowledge and hard facts that remain constant.
For absolute requests, they would follow the same approach as MySQL:
SELECT first_name FROM customers WHERE id = ‘p0ZoB1FwH6’ mysql
But deviates with more nuanced enquiries, with MySQL requiring some pre-thought on structure:
SELECT * FROM customer_attributes WHERE id = ‘p0ZoB1FwH6’ AND label LIKE “%preferences%” AND value LIKE “%invoicing%” mysql
And SynapseDB using a more fluid approach:
SELECT customers ID = ‘p0ZoB1FwH6’
PROMPT “What is their preference for invoicing?” synapsedb
“Our client, Fictive Innovations, prefers digital invoicing via email, sent monthly. They appreciate a clear, itemised breakdown of services rendered, alongside a summary of costs. Additionally, they’ve opted for a paperless approach, aligning with their eco-conscious ethos, and favour a seamless, secure, and straightforward online payment portal to settle invoices.” response
Designed for token-based environments
While token limits have increased vastly in a matter of months, the higher limits do come at a cost. The system is designed to be low cost, but retain strong performance and operate at various depths of context. This design strikes a balance within the limit, to provide the deepest knowledge available, while leaving room to converse and run your new query.
Use Case & Next Steps
This concept is now in the hands of the capable Forge Partnership development team to experiment with. It’s a thrilling yet demanding venture, and while the prototype is still on the drawing board, the potential applications are broad and promising:
- Healthcare: Streamlining patient data management.
- Content: Enhancing real-time collaborative editing.
- Customer: Elevating customer service experiences.
- Educational: Facilitating adaptive learning platforms.
- Scientific: Accelerating research data analysis.
- Smart: Enabling smarter home and city solutions.
- Supply Chain: Optimising logistic processes.
- Natural Language: Advancing language understanding.
- Social Media: Refining content moderation.
- Game: Boosting real-time game development.
And it doesn’t stop here. Any applications harnessing ChatGPT or similar LLMs could benefit, not to mention enterprise implementations improving InfoSec and safeguarding company assets.
The journey ahead is steep and jagged, but the view from the top promises to be worth it. If you’re interested in integrating with your application, or have the skills and find this topic intriguing, we’d love to have you on board. Get in touch, I’d love to hear from you and work on this collaboratively.
