This post is kind of funny and serious at the same time. I’ve always been a creative from producing “finrap” music eons ago to building a curbside app pre-pandemic (I dropped the ball on that one). Anyhoot, recently, I had finished writing my first 4 blog posts on FunDataMentals.com, and then asked my Whatsapp Group chat friends to please provide constructive feedback. One of them asked this question -- “what percentage of the posts were done by AI?” And that initially shocked and confused me.
But then I thought to myself -- “hmm, I can see why he or anyone for that matter may ask that question!". I actually ranted about the usage of LLMs in my LAP post. Anyway, shortly after, another friend in the group chat, whom I’ve also known for many, many years responded -- “...no way Clarence is plagiarizing or having AI create. Clarence is a deep thinker, academic, thought leader. Lol. Looking forward to reading the blog, bro”. As to whether he started reading it is an open question.
This is a growing concern for creatives that I’ve been noticing, from those in art & music to writing, that is, signaling authenticity in this day and age of mass AI-generated content. Even while I was a participant at the Recurse Center -- in my batch, at least -- there was a split of people who were pro- and anti-AI in the engineering process. One fellow Recurser always made it a point, during presentation days, to discuss trade-offs when working on his projects; he neither leaned strongly in one direction or another. I even pair-programmed with an individual who did not want to look up AI for any assistance. No vibe coding. Absolutamente nada. I appreciated this approach, as it forced first-principles thinking (the crux of my blog), and the related problem-solving grind.
This is also my approach when drafting blog posts. When I start writing, I draft, initially freewriting ideas (irrespective of logical coherence) to keep the momentum going. I throw any thoughts or insights, like a child throws spaghetti on a wall (does this really happen?), into my WhatsApp “me” chat, Google Docs (this post was exclusively done in Google Docs) or sometimes an email draft. I call this process, “intermittent writing”. And these intermittent writing sessions have no schedule -- when I have an idea whether it is when riding the subway or even having dinner, I quickly pull out my phone to take notes. I may construct a paragraph (or 2), and then later at night during quiet time, compile them as best as I can in a coherent fluid manner. Even so, up until now, all of my posts have been the products of multiple iterations each, not primarily because of the ideation phase, but because of the “polishing” delivery or presentation phase.
So how did I use LLMs in prior posts? They were used, mostly to capture typos (although I left one in intentionally to elucidate a point in my what-is-data-post), webdesign layout, KaTex, ASCII diagrams (though I’m now debating going back to old-school pictures of my hand-drawn/whiteboard diagrams), pictures, etc.
My strategy, however, has been and still is to own the creative process (the Fun part), and outsource the low-value grunt work. So yes, I delegate low-value grunt work to LLMs -- ChatGPT, Gemini, etc. Obviously, what is considered low-value grunt work depends on the individual. Case in point, designing a fun sockwear concept was a low-grunt task for me at the time of writing intp post. Don’t get me wrong -- it still took some time…a lot of time, in fact, close to an hour to get the right image. Here were some AI-generated outputs:
(A)
(B)
(C)
(D)
Which one do you like? Perhaps you like none. All good. I’ve settled on (A), anyway. The point is, if you’re a sock designer, this may fall in your realm of fun high-value creative expertise. For me, I just wanted to “quickly” get this post out, although carefully reviewing an LLM’s output is still time-consuming work. LLM verification is no walk in the park. Just look at (D) -- for the life of me, I couldn’t get Gemini to swap the left for right sock (without it transforming to C)!
Creativity is mine, Presentation I outsource (sort of)
“WTF is this trash!?!” (in a softer paraphrasing tone 🤫), I responded to an LLM output when I once experimented with the “creative” aspect of blog writing. It wasn’t connecting my ideas together I initially thought it would, even at times, biasing my ideas towards one or more textbook standard ways inconsistent with my own mental models. Unacceptable! It actually has slowed down my creative (or ideating) writing process, so much so that I completely abandoned LLMs for draft writing, only reserving their use for clean-up and presentation. More on this later.
So how to stay creative in this day and age? To answer, I look at this from the perspective of the person doing the creating, that is, I first ask the question -- “creative, relative to whom?” If relative to me, then I define creativity as being able to independently arrive at a conclusion, no matter how laborious the grind is. That is, no reliance on textbook (or similar) definitions and explanations to get there 100% of the way, no LLMs to give you the answer right away, no shortcuts! Even if the conclusion is not novel, the synthesis and the path may be (relative to the creative). And who knows -- developing a habit of independently focusing on the creative process, painstakingly navigating all the twists and turns, may someday yield novel solutions to existing or unsolved problems (relative to a much wider audience beyond yourself). History is not short on examples, one of my favorites being the brilliant Indian mathematician, Srinivasa Ramanujan, who in isolation, creatively both in process and conclusion, made significant contributions to several fields of mathematics including Number Theory.
Now, I love math. I even have a degree in it. But I’m no Terrence Tao. And, sure, I may not come up with some novel theorem or discovery in the field of math or computing. However, that doesn’t stop me from working on my own or existing problems. It doesn’t stop me from deriving (or attempting to with little assistance) already-known results in those fields. Case in point -- when I was in high school multivariable calculus class, my teacher once said “it is impossible to integrate a factorial”. I was like “huh, is it really?” That same day at home, I explored what would happen if I expanded out polynomials like x, x(x-1), x(x-2)(x-3), x(x-1)(x-2)(x-3), …” :
I explored the relationships between these coefficients, and thought I invented a new triangle with identity: F(n, k) = n x F(n-1,k-1) + F(n-1, k), where n and k are non-negative integers, n and k being the row and column numbers respectively.
Now, notice that P(x, k) = x(x-1)(x-2)...(x-(k-1)) = x!/(x-k)!, where P is a permutation, or number of ways to select groups of k objects from a distinct set of x elements. So I had a thought-experiment -- what if I try k = -1, -2, -3,.., etc. For example, P(x, -1) = x!/(x+1)! = 1/(x+1). And in general, P(x, k) = 1/[(x+1)(x+2)...(x+k)].
“Cool, so what next? Hmm, let me try adding these polynomial expressions, plugging in various non-negative integers for x, using my TI-80 (yes, I'm showing my age): … + 1/[(x+3)(x+2)(x+1)] + 1/[(x+2)(x+1)] + 1/(x+1) + 1 + x + x(x-1) + …”.
Now, recalling the original problem I was working on, I had asked myself, is this sum proportional to x!, that is, does this sum approximately equal or converge to Cx!, where C is a constant? And the answer is a resounding "yes" -- after plugging in various values for x and k, that C was Euler’s constant e = 2.718….
I was thrilled! I created a polynomial representation of a factorial over x where I can now integrate. I thought I was a genius for having discovered something unseen in mathematics…nope, not even close! It was not until many years later, after having access to more math literature, did I discover that these triangle coefficients were called Stirling Numbers of the First Kind. So yes, the factorial [function], technically the continuous version of it -- the Gamma function -- can be integrated.
So no, I did not discover anything for a larger audience beyond myself, but having discovered it for myself was still enriching. Had I encountered the Gamma function and Stirling numbers before I heard my teacher’s assertion, then maybe I wouldn’t have engaged my creative problem-solving and exploratory muscles for that particular problem. Was it a waste for me having gone through a process of re-inventing or re-discovering? Nope!
DO reinvent the wheel
We've all heard advice saying “don’t reinvent the wheel!”. I reject that (IF time is not a hard constraint)! There’s nothing wrong with reinventing the wheel (i.e., deriving from scratch), if you think it will help you invent (or debug) a better or different wheel, wing, robotic arm, or what have you. Insights gained via independently deriving existing results in one field can translate into discoveries and insights in others.
How do I do this in practice? I look at systems engineering or reverse-engineering, even conceptually. Here are some examples,
- The other day, I yelled out to my Alexa home device “Hey Alexa, bedroom Light One on!”, and then thought about how this actually works -- from sound pressure waves to my smart bulb turning on.
- I was vacuuming my apartment the other day, and then thought how would I conceptually build one?
- Oftentimes, at night, I stare into the sky, and think about how to conceptually engineer the universe (I have a whole philosophy on existence and the universe, based on my belief that the universe is “self-observing”).
For you, the system may be different. Maybe you prefer to conceptually engineer or reverse-engineer solutions in human behavior & psychology, medicine, sports, biomechanics, etc. Maybe even math?
I don’t do math anymore, though if I had infinite money, I’d probably go all in on math and theoretical physics, with bouts of hard engineering. And speaking of math, even when writing this post, another question came to my mind --
Given a blog post, what is the likelihood that it was AI-generated?
Many people would say “look for the [frequency distribution of] em-dashes”, among other patterns, like another buddy of mine in the same WhatsApp group chat. He literally said “I better not see any of those [em-dashes] in your blog”. And this is not the first time I've heard of this. I naturally use double-dashes in practice (when typing) in lieu of em-dashes, but em-dashes still have their place in grammar. Perhaps, it’s over-used so much in LLMs, that some people squirm when they see it in a blog post or a pitch deck asking to fund their next unicorn startup.
More specifically, I think this question should be bifurcated -- 1) "what is the likelihood that [the presentation/delivery - grammar, diagrams, etc.] was AI-generated?" and 2) "what is the likelihood that [the content] was AI-generated, given your belief that the delivery was AI-generated?" The latter is what I fear.
Why Do I Care, Why Do I Fear
Well..there’s a practical reason. I’m broke, and no -- not regular class broke, not even high class broke:
, but low-class Broqué. I recently got laid off. And I figured -- “hmm, maybe me blogging all about my first-principles and bridge-principles ideas I’ve been keeping to myself, would differentiate me from other software engineering candidates”. My thinking was that tech hiring managers and recruiters not only want to see evidence of concrete engineering artifacts and use of AI tooling, but also evidence of deep understanding, creative thinking and domain-agnostic problem-solving that could translate to cleverly solving known and unseen problems. Has my approach been working so far? Nope, not even a bit! So I suppose I’ll give it some more time, otherwise I’ll resume playing poker to pay rent, and maybe, just maybe, once in a while, publish a new blog post into the void. All jokes aside. This market is rough. It takes skills, persistence, LOTS of networking, and a bit of luck to gain income. But I still have to ask -- has this AI boom created a low signal-to-noise ratio for you creatives, regardless of how you want to use your authenticity?
Conclusion
Authenticity is at least two-fold -- one in Content and one in Delivery. I’ve been focusing more on the former, and losing my ‘voice’ and personality in the second. Perhaps, if I can train an LLM to preserve my voice in delivery, then I could outsource the latter.
What are your thoughts on using AI in your “creative” writing process?
To em-dash or not to em-dash? That is the prompt.
[PS, no AI was used to generate this post, not in ideation, and not in refinement].