Shelby's Blog

Step-Based Cascading Prompts: Deterministic Signals from the LLM Vibe Space

A new technique for enforcing LLM behavior conformance in decision making workflows.

Text Segmentation for RAG and the Shaky Foundation of AI

Text segmentation and it's common shortcoming in RAG, LLMs, and AI.

Configuring persistent power limits on Nvidia GPU's in linux

A simple step-by-step guide and script for automating power limits at boot using systemd.

Retrieval is all you need (Part 2): RAG building blocks and stacks

Cataloging the SotA technologies available and permutations of stacks for retrieval pipelines in Gen AI. Part 2 is a continuously updated collection of technologies and workflows for RAG.

Retrieval is all you need (Part 1): A brief intro to text retrieval in RAG

Cataloging the SotA technologies available and permutations of stacks for retrieval pipelines in Gen AI. Part 1 is a brief intro to RAG, and an overview of retrieval techniques and technologies.

Tested: Mixtral 8x7b vs. GPT-4 for boolean classification

LLMs, non-programmatic computing, and logit bias.

Exploring the syntax we use to describe LLMs and a discussion on how to use LLMs to make business logic decisions in software with logit bias.

Template for creating new posts

This post is used to generate new posts. It also contains instructions for adding posts.