My name is Mike and I like AI (And sharing with Collabry and our community)

By Mike Golaszewski, Senior Consultant

How it started

I’ve been fascinated with computers since around 1980, when an 11-year-old me typed the word “RUN” into a little Tandy computer and was surprised by a game of hangman. I had accompanied my dad to the local Radio Shack store, and — bored — wandered over to play with the calculators. The PC-1 “calculator” was the most interesting model on display because it had letters! In the corner of its two-line LCD screen was a little tiny word that said “RUN,” and so I just typed what I saw and pressed the enter button. The magic that this moment unlocked has defined my life ever since. The idea that I could get this little machine to do anything I wanted was intoxicating to a small, bookish kid, and within a few months I had convinced my parents to get me an Apple computer. From there it was off to the races.

How it’s going

I was reminded of this moment back in late ’22, the very first time I used ChatGPT. Artificial intelligence wasn’t new to me — I’d advised an AI-based travel startup as early as 2016 — but I was once again surrounded by an energy that seemed almost magical. The idea that now anybody could get a machine to do anything they wanted seemed unbelievably powerful. I was captivated all over again. Since then, I’ve spent a lot of time experimenting with several large language models (LLMs) to learn what they are capable of, what they’re not really good at, and what makes them tick. This research has included adversarial prompt engineering (getting the model to do things it’s not supposed to do), chain-of-thought prompting (refining inputs to discrete steps), response shaping (helping to define and control model output), meta-cognition, reflective planning, and agent design.

What this means for financial services

Now, as a Senior Consultant for Collabry as well as other clients in the financial services space, I think about the immediate implications of AI in this sector. For now, most of the hype is still just that: hype. The real, most immediate impact of LLMs is in high impact but unglamorous areas like meeting transcription, complex document review, and regulatory compliance. Outside of that, the other interesting concept to me is using these models to explain a client’s investments and returns in clear, simple language. 

Paradox: those who stand to benefit most

Another finding is somewhat unintuitive but something I long suspected: The professionals in the best position to maximize the value of these LLMs are those whom these models most seem to threaten: writers, editors, and other linguistic creatives. Why? Getting these models to perform is highly reliant on descriptive, structured, and precise language — skills that anybody who has had to hold a professional pen for any length of time has trained on, tirelessly. ChatGPT, Gemini, DeepSeek, and other LLMs may have been built by mathematicians, data scientists, and computer engineers, but only a well-trained writer can make these models truly sing.

Stay tuned

Over the next month, I’ll be hosting a four-part series for my colleagues at Collabry on how AI works, and how writers and other creatives can leverage these models instead of competing with them. I’m looking forward to sharing some of these thoughts here and in a Community Workshop in March. I welcome your thoughts!