I am admittedly fascinated with large language models. I’ve been enamored with the idea of automated assistants since before the days of Microsoft Bob. The notion of a searchable, integrated, digital-life-spanning, automatable solution appealed to me. I can hear some of you thinking, “Why did it appeal to you, Andy?” I cannot articulate a rational answer to that question.

I Use ChatGPT

Why? Because a searchable, integrated, digital-life-spanning, automatable solution still appeals to me.

How I Use ChatGPT

I have been a ChatGPT subscriber for a while now. I initially subscribed to learn more about how ChatGPT works. I remain a subscriber because I find ChatGPT to be a productivity accelerator and force multiplier.

I use ChatGPT in a couple / three ways these days (06 Mar 2024).

1. I use ChatGPT to accelerate document generation. The documents include product documentation for Data Integration Lifecycle Management Suite products I maintain and sell.

Generative AI, in general, is good at generating a lot of text. Someone once said, “There are no good writers. There are only good editors.” The quote – or a paraphrase – is attributed to a few different people.

Later in the post, I’ll share one way to build a custom GPT to help format code when developing software.

I did an experiment a while back where I genericized a command file and asked ChatGPT to convert the contents to PowerShell. The resulting PowerShell “worked” inasmuch as the execution did not fail. Further examination revealed the PowerShell also did not do what the original command file did. I’ve learned ChatGPT and other LLMs do not (yet^*) possess the ability to generate code that Just Works.

^{* Will LLMs generate code that Just Works in the future? I believe they will.}

For generating a lot of verbiage suitable for a human-in-the-loop to edit, generative AI / LLMs are awesome. I generate and edit.

2. I use custom GPTs to accelerate tasks. There are a number – over 100 at the time of this writing – of specialized GPTs that are trained to perform specific tasks:

I’ve used a few GPTs and experienced enough success to explore more.¹
I confess I’ve not enjoyed much success with image generation, but I attribute the misfires to my prompt engineering skills.

3. I’ve built a handful of custom GPTs. Custom GPTs allow users to configure personal^* chatbots. For example, you could base a GPT on your favorite coach or someone famous. You could add more people to the mix for a blend of context. The personalities may be fictional characters from books or movies. LLMs are good – and getting better – at mimicking the communication styles of people they “know.”
   ^{* May not actually be private.}
   To learn how to build a custom GPT, see my post titled One Way to Build a Custom GPT.

Conclusion and More Information

The options described in this blog post are not the only ways to work with large language models.

In 2023, I founded Enterprise Data & AI to help enterprises implement solutions based on large language models. Data engineering remains an important step, vital to providing, curating, and cleansing the data used to train LLMs – so I’m still busy at Enterprise Data & Analytics! Click here to learn more about entdn.ai.

^{1 I did not use GPTs in generating or editing this post. I have used GPTs in the past, in the manner I described earlier for post content and outlines, and then I’ve edited; correcting outright mistakes and rewording to better express my sentiments – and I will continue to do so.}