Of Categories

Categorization is the process by which the possible contents of thought are first organized prior to cognition. There is an infinite diversity of content in the world that requires ordering to be represented and understood. The contents of cognition when lent to disorder and unstructured representation fail to communicate truth, leading to crude understanding and confusion. Cognition rests upon the synthesis of raw content and structured categories

This human process of ordering and structuring has been forgotten when it comes to the development of machine reasoning. Discourse around LLMs displays a fascination and a mysticism, attributing undue powers to LLMs with the expectation that all existent problems with the technology shall fade due to increased processing powers. This progress of machinic reasoning supposedly has no conceptual bounds in the conquest of truth, merely technical ones; yet, the actual application of LLMs to the world has found itself mired in the infinite diversity of possible truths to be attained from the world. Lacking categorical structuring, machinic reasoning fails at cognizing the contents of the world. Synthesis of the world becomes impossible to LLMs. It is odd that this basic insight of categorical structuring has been forgotten for categories themselves are technologies in the truest sense - that which orders the world

We see remnants of categorical structuring strewn throughout the current landscape of LLM databases - chunking, vectorization, knowledge graphs and more play a role. What has lacked though has been the maximization of the recognition that it is structured categories and not the progress of machinic reasoning that will bring order to the infinite diversity of content which is the world. To accept this is to realign the current trajectory of LLMs away from the mysticism of research labs and into the practical world of real business applications

Gestell marks a fundamental transition of approach to machine reasoning. Rather than the belief that the unbounded content of the world will be intelligible to LLMs, Gestell embraces the understanding that structured categories take part in the synthesis of the content of the world. Gestell proposes that a singular type of categorizing cannot structure everything, rather given the multiplicity of use-cases of LLMs, there needs to be a similar diversity of categorization. Gestell is built upon categories, and lets you structure your data with them

What is a Category?

A category is a guidepost. Through the entire data structuring process - from Enframing to Disclosure - categories determine the exact contents of data that are structured, the relations between frames of information and the presentation of these frames. Lacking categorization, structuring processes are naive and fail to capture the necessary portions of content that would be required in complex reasoning. Categorization facilitates the maximization of structure through instructing what order looks like in the data. When an LLM comes to traverse the possible contents of its‘ cognition, that which is primary to the truth is presented first via categorical structuring. Categories serve as the rules for structuring the world, and thus serve to present the world in an intelligible manner

With Gestell, categories are simple to create - use your natural language. State what you need Gestell to do with the data, and Gestell will structure your data accordingly. Once structured this data will actually be intelligible, at scale and accurately, to an LLM. These categories can be as complex as your use-case requires, enabling deep and reflective reasoning where traditional search processes would fail. Categories are immediately deployable to structure your data, no fine tuning required, allowing live production-level scale to search-based reasoning in moments

In a business context, what does this look like?

A category is a specific and high-volume workflow that you need for data structuring (Enframing) to best complement search-based reasoning (Disclosure). Once your data has been Enframed, we suggest adding categories on top of it for specific workflows you want to accomplish

A Feature Extraction Category will be able to extract concepts or ideas from a database in a far more effective and efficient manner than trying to do normal search. Giving structured outputs of the exact features from your data

You can also introduce categories to overlay specific concepts upon your data structuring. If you are building a knowledge base of possible contracts for an M&A Legal assistant, you can implement categories to structure search-based workflows for things like Reps and Warranties, Termination / Breakup Fees and any other complex or high volume uses. Or, if you are building a customer support agent, you can overlay categories of specific troubleshooting workflows or references to documentation

Each category rests on-top of your Enframed data, enabling these high-volume workflows to achieve even greater precision and efficiency. If you have struggled with traditional retrieval processes, we encourage you to instead try introducing Categories as data workflows. Categories are live now on Gestell