I was looking for a good dataset to use for comparing different models in a RAG application when I found this post on Reddit. It compares a bunch of models on a collection of questions over a set of documents provided by H2O.ai.

I wasn't super interested in the benchmark, but the files (mostly pdfs, one mp3, jpg, other file types) interested me for use in my own testing. This short post shows how to get them using the scripts provided by h2o.ai.

This short post shows how to use the org-clone-subtree-with-time-shift command to make org headings for recurring tasks. I was recently trying to add a five-week class to my org agenda and I didn't want to manually create each heading and add or modify the timestamp. This approach made it very easy.

Today I was customizing the appearance of org files displayed with org-tree-slide. In particular, I wanted to increase the font size and start Olivetti mode whenever I started org-tree-slide-mode and return everything to normal when I was done. This, I quickly discovered, required the use of hooks. Hooks are not especially complicated, but they are useful and worth taking a few minutes to understand. This post will cover the basics of working with hooks in emacs.

The wonderful chatgpt-shell package by Xenodium lets you interact with the gpt-3.5 and gpt-4 APIs in emacs via a handy shell built on top of comint-mode. It also integrates well with org-mode.

I find that I tend to re-use a few prompt patterns for specific tasks. Yasnippet provides a great way to create prompt templates made up of some fixed component with placeholders for user input. I can easily insert these prompt templates when working with chatgpt-shell to gain easy access to reusable, task-specific prompts. This post describes how to start using Yasnippet for prompt templates for use with chatgpt-shell.

Since this start of this year, I've been working on and writing about AI tools for working with Postgres databases. Most of this work has involved finding different ways to integrate ChatGPT (and previously Codex) with other tools and workflows. I wanted to collect and share some of that writing here, as it's related to a lot of the other things I write about on my personal blog.

One of the things that makes working with the ChatGPT API a little different from working with, e.g., the davinci-text-003 model api is the need to maintain the history of a given chat session. A Julia Struct containing the chat history, coupled with a function that acts on that Struct, provides a good way to work with the ChatGPT API.

For the basics of working with the ChatGPT API, check out part 1.