I had what I thought would be a great addition to the Daily Blogroll. There’s been many times where bloggers talk about the same subject, like World of Warcraft or EVE Online, but just not on the same day. What if I could have a broader index of recent posts on the same subject so that someone interested in the subject could just see, in one place, all the recent blogs on that subject?
That became a really interesting exploration of just how that could be accomplished. I would have AI write the summary, of course — all of the Daily Blogroll is AI generated — but gathering all that data would be a struggle.
Step one was to figure out the categories. I run each day’s summary through ChatGPT to generate hashtags, and I figured those hashtags would form the base categories. Next, I already read everyone’s RSS feeds to make the Daily Blogroll. Instead of just using and discarding the posts, I’d save them in a database, and use GPT to save which categories each post belonged to.
Then I’d see which posts that had not been part of a deep dive had common categories that would pass a threshold of number of posts on a topic, read the full text of all those posts, and send it all to GPT to write an article, with links, that brought all this together.
Problem was, the AI writing stinks. I hated it. I kept trying to adjust the prompt to make it at least slightly readable, but I failed. All the trips to the AI to get to the “deep dive” were pretty expensive and the result was unreadable trash. From an economic sense, it made no sense. From a creative sense, it was just further tarnishing my reputation as someone who uses AI on a day-to-day basis.
So, as of tomorrow, no more “deep dives”. I still think there’s value in there somewhere, but I don’t think the current state of AI can produce readable prose.
It was a fun experiment, even though it didn’t really work out like I’d hoped.






Leave a Reply