The archive for this weblog has been set to private for any posts published before 2025. That means 2,382 posts are not set to private. Only 51 remain public before this one goes live. Intellectually I know that some of those older posts are part of archives or were scrapped for training. My writing style is uniquely mine and I don’t think it can be truly mimicked, it can only be reasonably replicated. Right now the current generation of models are able to take a corpus of my writing and spit out fairly convincing prose written in my style.
Recently, I have been wondering if I should try to refocus my research note based writing style each week. The Lindahl Letter is quickly approaching 200 Substack posts. Based on my tracking system next week is the 193th missive, but the 199th overall post to the Substack system. Any way you slice it we are quickly approaching a major publication milestone of 200 blocks of content being published. All of that is happening over on the www.nelsx.com domain after that Substack got switched over to a custom domain for hosting. Now that we are reaching that major milestone I started to consider if a best of post was appropriate or if maybe just maybe I should change directions a little bit and do something different with my weekly writing efforts.
Each weekend I sit down at the start of the day and begin to work on research and writing efforts. During that time I simply attempt to deeply focus on whatever topic is being considered and write down my research notes into a consolidated overview with my thoughts, predictions, and attempt to summarize the material. I did train a model to actually produce the same type of content that I would produce in seconds instead of hours, but that is not a very fulfilling process or really very interesting for me as a writer. I actually think at some point somebody is going to make an organic writing only search engine. People are going to get tired of the never ending stream of AI generated slop.
One of the potential areas that I may spend some time researching would include AI use cases with a verifiable ROI. These research notes would be like case studies, but effectively shorter. Gathering together a bunch of these different use cases would be the basis of a larger analysis of this area. Another part of that might be trying to locate the 10 most expensive use cases that failed or at least did not realize a quantifiable benefit.
Maybe deeper digital histories from people before the rise of AI slop will end up being more valuable. Unfortunately, a lot of these deeper histories are starting to disappear. Most of the older computer builds that might have housed these archives are slowly falling out of usage.