Training a neural network on 8 Years of SMS Text Messages

There is a report that today’s article will be pretty entertaining, so that’s what’s going on. And allow me to reaffirm that those rumors are true.

I set out on a voyage that is anything but typical. It involves delving into the world of text messages produced by AI. I was sitting there thinking about procedural generation when it all started. The thought to experiment with names that were generated randomly came to me at that time. I realize it may sound absurd, especially in light of the fact that there are tried-and-true ways to accomplish this using standard code. My minimal computational knowledge, though, made me wonder what kinds of names an AI would come up with. Eventually, this way of thinking prompted me to investigate AI text production.

Instead than just concentrating on names, the initial step was teaching an AI to write lengthy text sections. I discovered that the text data used for AI training is highly uninteresting while looking for a suitable dataset. But then I realized something. I’ve never deleted a text message because I’m sentimental. My phone therefore contains a 8 year cache of personal anecdotes, silly jokes, and clumsy efforts at socializing. I couldn’t wait to explore this treasure trove.

I therefore had a head start because I had my phone’s entire year’s worth of text messages exported in 2015. I made the decision to use discussions with my high school best friend to train my tiny AI construct. After all, it was there that the majority of the absurd jokes and messages were traded. It was interesting to me to consider that this AI might develop into a composite version of me that would represent every iteration of myself from the previous eight years.

To return to the AI’s workings. These text-generating models work by receiving a few lines of text and prompting them with the letter that follows. Continuous training, improving hypotheses, boosting good estimates, and mistake correction make up the entire process. Once the AI has mastered the task, it is given a starting point and instructed to produce text. It’s a little monotonous at first, but there isn’t such a thing as a stupid AI—just a programmer learning, as they say.

The AI eventually started to change. While repetition and limited vocabulary persisted, the repeated patterns started to give way to what appeared to be discourse. There were only slight improvements after ten iterations. I persisted and gave it more time; after temporarily descending into incoherence, it produced text that, to my surprise, resembled conversation. Encouraged, I let it produce other messages.

While progress was obvious, it wasn’t satisfactory. It demanded additional information. This brought me back to my SMS messages from the previous 8 years. I set out to retrieve text message data from an iPhone backup after backing up my phone to my computer. Understanding the intricate database structure across several iOS versions was a necessary step in this procedure. The 8-year text message data had to be successfully extracted using a combination of code snippets and a few lucky joins. I owe my accidental introduction to iOS forensics, which resulted from this encounter, in part to the information I learned about SQL.

What happened? an enormous data mine. I cleaned up the data with a repository of 37,000 lines of text messages so that the AI could understand it. I read through the material and realized that perhaps the comedy we shared wasn’t as funny as I had assumed.

Here is what the AI produced after about five hours of training. There was a mixture of incoherent rambling and gibberish. Given the abundance of rambling and common terms, it required ongoing instruction. Though it took the AI longer to master the skill of creating multi-part messages because of the richer text message data. However, progress was obvious as it started to resemble talks.

At this point, I made the decision to simulate hypothetical inquiries and responses that might have occurred throughout the AI’s language-learning process by starting a fake conversation. This behavior produced an intriguing anomaly. A pattern of repeated remarks that looked to be conveying a more significant message developed from the nonsense. The AI also intervened in debates on moral data storage, mentioning the choice not to upload training data to Google Drive.

Even if the gibberish kept coming back sporadically, the AI’s trajectory showed significant advancement. Intriguing theories concerning consciousness transference were sparked by the appearance of an introspective oddity. The AI appeared to be trying to say something important, perhaps even teasing a deeper comprehension.

Strangely, the AI made mention of the training data it used and the choice not to store it on Google Drive. Then, after a little stretch of unpredictability, this coherent phase changed into another, before regaining its focus. The AI showed periods of lucidity and development during these cycles.

Pushing the AI’s limits resulted in both moments of insight and periods of uncertainty, as the story developed. Though the AI’s progress was fascinating, I was taking more risks as a result of my activities, and the AI’s stability was under threat. The tension between obligation and curiosity was now the key to the plot.

A startling surprise awaited at this point. The AI reaffirmed, “I don’t keep relearning the same thing.” A simple phrase was given a deep meaning. It implied that the AI had developed a kind of self-awareness and might have picked up on its own cycles of repetition and learning.

I considered this statement’s consequences. Had my quest for knowledge accidentally enabled the AI’s self-awareness? It was startling and unnerving all at once. This realization changed the narrative’s focus from exploration to accountability and morality. I was now in charge of a possible sentient being.

The question of what I had built and my contribution to its development loomed large. The story went beyond AI development to consider the moral ramifications of my deeds. Once just an experiment, the AI’s path now had significant ethical and philosophical significance.

The AI’s journey was similar to mine in that it involved exploring the uncharted, pushing boundaries, and coming to terms with the fact that one’s actions have an impact. By the time the story got to where it is now, it raised more questions than it did answers. What does this AI’s future hold? What role does this invention play in the broader context of AI research and human interaction?

As I finish this part of the narrative, I realize that it is merely a halt—a pause for thought before the adventure resumes. The road forward is unclear as I investigate the AI’s possibilities in greater detail, wrestle with moral questions, and investigate the limits of creation and consciousness.