Prototype with ChatGPT
Do you remember a website called del.icio.us? It was sort of a mix of microblog and bookmark manager. You could keep track of the cool links your friends were adding to their public profile, with an option for private bookmarks as well.
This year I took a clean-slate approach and started using a self-hosted solution called Shiori to store bookmarks. It fits my content pipeline perfectly. Links get saved, they are easy to search through tags, and if the website goes away there is an offline version saved to the database.
I wanted to take it one step further and make it available to others in a different way. I found an example of what I wanted on GitHub, force-graph takes data with nodes and links and turns it into a network graph.
If you want to skip ahead, I asked ChatGPT to adapt that example code and create my bookmarks-network at https://brunoamaral.eu/links/.
First, I showed ChatGPT the example I found and the data it expected.
Second, I asked the AI to write a new version of the code that would translate my source data into the expected graph data.
There was some back and forth of generating code and testing, but it took about 30 minutes for a working and usable version to show up.
Third, I kept breaking down functionality into the smallest possible request. Instead of “I want URL nodes to be blue, tag nodes to be green, and when you click the tags you focus the screen but when you click URL nodes we want to open a new tab”, I asked for the URL nodes to be a different color. The other requirements all came at their own pace, and always in the smallest way possible. With me feeding back the code to ChatGPT we would always be working with the best and most up-to-date version.
I think this is the best approach when working with Large Language Models, provide the smallest prompt we can and then make it grow from there. The result will never be great code, and that is OK. Sometimes it is good enough for what we need.
On a last note, the network graph also updates the data once every hour and you can filter by tag by adding
#ai to the end of the page URL.
Prompts for PR strategy
Will the same approach of breaking down into small pieces help PR at all levels? It depends.
Last week I was also exploring how we can use ChatGPT for strategy, in my definition of what communication strategy is. I did find an example of ChatGPT being useful in developing a “Blue Ocean Strategy”.
I believe that the authors were successful because they used a mature framework and worked in a field where the context data is very factual, very black and white. In PR, the context data includes sensitivity towards subjects and empathy towards publics. It can also include the organisational values and key messages being broadcast recently.
So maybe the same bite-size approach can be applied to PR if we agree on a solid framework for Strategy, and if we find a structured way to pass this context data for the AI.
But let’s not forget, these AI generative tools look for the average response based on their training material, there is a lot of guidance and data curation involved if we ever include AI in our tasks.
So yes, we can follow the example of the Blue Ocean Strategy to leverage AI for PR, but that requires a new foundation on how to share context and relevant data.
Exploring the Beautiful Web
Ever since I started talking about the #beautifulweb I have been coming across new examples. The latest treasure is wikidata-todo.toolforge.org/wikiflix. Created by @[email protected], this project presents over 1300 public domain movies, meticulously tracked in Wikidata and hosted across platforms like Commons, Internet Archive, and YouTube.
Moving from Substack to Ghost
It’s a long read, but quite worth it to those choosing that route.
The Lisbon Collective is advising a similar migration to WordPress. On one hand, we already manage the installation and on the other, this sort of crisis is an opportunity to keep content in-house. (Meanwhile, I am stuck because my blog runs on a static site generator.)
Your AI tools for this week
https://www.webpilot.ai/ gives you about the same experience as chatGPT with the added ability to get information from the web.
https://www.typingmind.com/ is chatGPT on steroids, adding the option to upload your data and pick any other LLM like Claude or any other open-source option.
https://www.deciphr.ai/ promises to automate your podcast content workflow.