How do we fix the “Too Many Newsletters” problem?

“I subscribe to so many Substacks, and I read zero of them,” quipped Foursquare co-founder Dennis Crowley on one of his social accounts (I honestly can’t recall which one). This perfectly encapsulates the issue with the newsletter ecosystem.

Email newsletters were intended to address the problems of “RSS feeds” by consolidating all the information in your inbox. They aimed to streamline reading and information consumption. Newsletters promised to help writers stand out amid the noise.

However, we now face the same issues that plagued blogs— an overwhelming abundance. With platforms like Substack, Beehiive, Ghost, and Buttondown making it so easy to start a newsletter, it’s no wonder we’re inundated with “newsletters.” Just as blogs transitioned from being lauded as efficient to becoming tools for content marketing, newsletters have similarly become synonymous with content marketing.

The frequency with which people add me to their Substack email newsletters weekly is remarkable. What many don’t recognize is that, in doing so, they induct me into Substack’s vast spam mechanism. There seems to be no legitimate way to remove my email address from their database. The “unsubscribe” option feels illusory. Consequently, I’ve taken to blocking Substack emails entirely.

Instead, I choose to subscribe to handpicked email newsletters using a disposable email address provided by Feedbin, which serves as my primary feed reader application. I’m a fan of Feedbin and believe its service is worth the annual fee. (See my review.) Much like Dennis, my inbox also houses a plethora of newsletters—around a hundred, in fact. And there are new ones that are so good that they have to go on the list. 

AI TO THE RESCUE

Like everyone else in Silicon Valley, I’ve been pondering the best way to utilize the new software layer. With the advent of machine learning and large language models, there must be a smarter method to help me navigate my expanding list of newsletters. This was at the forefront of my mind a few days ago when I caught up with a friend who has been in Silicon Valley slightly longer than I have.

I first met Nitin in the early aughts as I was exploring what would come after relational databases. Nitin is somewhat of a polymath and a maverick in the database world. He served as my informal chief technology officer when I transitioned my blog into a publishing company. Throughout the years, he’s been a consistent sounding board whenever I’ve sought to envision the future from a software perspective. Six years ago, he elucidated the impending wave of “machine learning” that would assist in monitoring, augmenting, and eventually generating code. Currently, he’s a consultant — and I’d highly recommend him to anyone looking to define their post-AI software strategy. (Nitin on LinkedIn/ Twitter.)

I presented my “too many email newsletters” dilemma to him and pondered if we might employ some of the latest language models and natural language processing capabilities to devise a solution. My MacStudio could be the perfect test bed for this solution. It boasts ample computing resources, is replete with GPU power, storage, and memory, and is connected to a one-gigabit ethernet internet connection.

After delineating the application’s flow, we brainstormed how to get it started. First, I created a separate folder on Feedbin for “Substack-only” newsletters. This simplification is useful because the Substack email newsletter format is remarkably consistent. Moreover, it ensures we don’t bombard an early version of the app with 100 newsletters. The RSS output from this folder is processed by the app, which utilizes open-source software and LLM models to summarize the content of all the newsletters.

The interface is still rudimentary, but the objective is to train the models to a point of personalization. For instance, I aim to refine the app to discern the frequency of a newsletter and how often I engage with it. Such insights would allow the system to prioritize what content is presented to me. It’s a significant milestone to achieve — step one. Ultimately, I envision the app summarizing content tailored to my preferences and delivering it directly to my inbox.

Who knows? We might expand its capabilities to encompass all RSS feeds and other sources and then unveil it for public use. It will be open source, naturally. Stay tuned for the next phase of this experiment.

Hey Dens, we got you!

August 27, 2023. San Francisco