Writing with AI
Introduction
Being semi-retired (because I can’t afford to be fully retired) leaves me with lots of time on my hands. However, I was one of those children teachers hated because I always had my line of questioning that often did not align with the lesson plan. Formal education and I did not usually mesh well (dyslexia and ADD were just some of my undiscovered superpowers). Curiosity was my superpower in a time (1960–1980) when conformity was the rule of the day. It’s no wonder I gravitated towards Woodstock’s rebellious element and whatnot.
Discussions with friends and strangers would often devolve into verbal conflicts over promoting a particular point of view, and rarely did I find someone who would engage in dialogue for the purpose of mutual discovery. Now, I have always been a technology-focused sort, not a fully qualified nerd. My dyslexia brain would not allow me to focus on the minute details needed to manifest abstract ideas into computer code. I also had issues formulating ideas into engaging, meaningful, and coherent writing. But that same dyslexia brain could grasp large-scale concepts easily; I recall conversing with my dad about quantum mechanics when I was 6. He worked on the Manhattan Project and was impressed with my curiosity but depressed with my lack of focus and academic discipline. One of those people I could converse with for the purpose of mutual discovery.
Dad passed away at the ripe old age of 53 (likely from radiation-induced cancer), and I was very occupied with trying to make my way in the world. He started my on a college career, a miracle in itself because my scholastic life up to that point was abysmal. Shortly after that, he passed away, and if I was going to finish, I needed to supply both motivation and finance. Much to my surprise I managed to succeed with that.
Fast forward 2024 and the tech twerp in me is curious to see what all the hubuh about ChatGPT is all about. I had worked with Expert Systems in the late 90s, back when that technology was called “AI.” Neural nets were just emerging and not widely adopted by businesses. I was surprised when the response to my prompt resembled a coherent string of text. Far more coherent than Eliza, an early program that influenced later LLMs developed in the mid-1960s by Joseph Weizenbaum at MIT.
At first, I found this neat trick entertaining and asked questions to confuse it and generate absurd responses; then, I found that when combined with detailed information searches and source confirmations, it turbocharged my research. I could ask stupid questions without the embarrassment of speaking with a human. And rarely did an LLM seem to have an emotional bias. A training bias was apparent, but some persistent questioning and challenges would eventually surface something akin to reality.
Due to my status as “retired” (pro-tip: your curiosity does not retire), I spent a substantial part of this time focusing on several LLMs, Claud, Perpelexity, Leonardo, and Deepseek, as a writing assistant. Claud allowed me to bypass much of the friction associated with writing due to my Dyslexia, and the ability to drill down on my line of questioning without anothers bias agenda intruding into the dialogue helped keep my ADD at bay. Then there is the whole being 70+years old, all those dysfunctional brain bugs just seemed to fade away. The bottom line is that writing with AI assistance was fun! I was never one who desired to be a word artist; I merely wanted to get my ideas across to someone else. I heard once something that rang a bell for me: young men work to make a living, old men work to make a difference. I want my ideas to outlive me; writing might accomplish that. At best, it keeps me off the street and out of trouble.
In this document, I want to focus on the AI-assisted journey of writing an article. Full disclosure: I use Grammarly because mi splleeng is not goood, typping also not well And my grammar is worser.
This article was written from an outline provided by DeepSeek and Anthropics Claud. I like to churn the same or similar questions/prompts between AIs to expand perspective. The actual writing (for better or worse) is all me, more information than art.
Foundation
I have found that a back-and-forth with an AI assistant with no agenda but to respond to my thoughts is similar to what is referred to as the Socratic method of learning. With each response from my AI assistant, I need to carefully read the results and further question them, feriting out uncertainties and ambiguities or inconsistencies with counterarguments or clarifications.
All the while, I am moving toward a cognitive framework similar to the way a sculptor removes material to get to a finished idea. This usually involves several iterations and summations to get a rough outline. This approach lacks the component of “Wisdom” that a learned human participant would bring to the discussion. But it also allows me to follow my tangent of thinking unencumbered by anothers emotional or intellectual bias.
A typical Article
Long ago, I developed a dislike of traditional media not enough facts and too much opinion. Opinions are like assholes everyone has one but I am very selective with learning about either. Opinion: Do you know the difference between a BMW and a pig? Never mind, or Google it, moving on.
I start by looking at a YouTube video by someone whose opinions I respect because they have a history of questions and answers that, after much critical review, I have come to respect. Generally, they are far more advanced in their domain than I am, so I am left with language and concepts that I don’t readily grasp. I then gather the transcript from TouTube into a text file and feed it to Claud or Perplexity and start asking questions.
Then there are the things that the YouTube discussion did not cover that have piqued my interest, so I start down that road often, back and forth for numerous iterations, until I have flushed out the obvious BS (AI bias and hallucinations) from reality. This often results in voluminous rouphage that needs to be condensed into something that is remotely interesting and engaging to a reader. At its essence, the conceptual framework is my rough intellectual sculpture; AI is the chisel.
Now starts the hard part: reading and re-reading to shape the rough edges into something that more closely resembles my finished idea.
My AI-assisted outline says I should provide 1–2 examples of these conversations. This is where my human brain says, “Are you crazy? Who would want to wade through 90 pages of chatter between an AI and me? You can do this yourself. Dump a document of interest into AI and ask a basic question or request a summary. Next, clarify any language that seems out of your understanding with respect to the subject. Now, in the same discussion thread, follow a line of questioning that is in line with your thoughts around why you chose this document to discuss with AI.
Conclusion
Note: Some chat tools can query the Internet, and some only rely on their training data set, which is limited to a particular time frame. I use a paid version of all the tools I mentioned above. Free versions might not be as robust. A pro (really expensive) version may be better, but I am retired and on a limited budget. I find that DeepSeek does not want to discuss things that may upset the CCP and will quickly want to change the subject. On the upside, it does supply a thought process before providing results, and I find following that interesting and thought-provoking. “Research” mode, while slow, gathers more information and provides citations.
This is how it works for my writing, but even if you adhere to one of my favorite axioms:
The illiterate of the 21st century will not be those who cannot read and write, but those who cannot learn, unlearn, and relearn. — Alvin Toffler
Whatever your area of learning you may find these assistants helpful little buggers, experts have shown us how these AIs have chewed through data and provided insights that would have taken human brains years to do.
It is a technology that will only get better with time. Unlike programs like Lotus 123 or Excel, it will improve exponentially, so strap in for a wild ride.