At Econometricus, we think that one of the paths forward for AI (LLM Realm) is the LLM-Less AI Software Application. That is, software that leverages Natural Language Processing without the LLM’s computational-expensive and feature-overrated dependencies.

There are a number of benefits to this new concept. At first glance we see,

1. Compliance regulations constraints whereby clients must manage and keep their data locally and secured.
2. Privacy and secrecy where trade-secrets are never-ever expose.
3. Lower Computational and Financial Costs where clients achieve more with less.
4. Reduced model hallucinations by implementing laser-focus tasks and content.
5. Tailored Solutions that Maximize Adoption.

The following video illustrates an implementation of this new concept of LLM-less software:

The strategy is as simple as agreeing to a syntax for English-prompting the software. If you noticed in the video, all we do is to constraint the prompts to a three components prompt syntax:

1. Call operator.
2. Call data involved in the operation to perform.
3. Call the orientation in which the operator execute: row-wise, column-wise, and/or pivot.

Developers had for years the toolkit necessary to handle effectively language/unstructured data, even before OpenAI’s release of ChatGPT. However, we fell short in the prediction tasks of language-related analytics. The trick moving forward is not only to leverage LLM wrappers. We must also deploy the full breadth of NLP tools we have understood for years.