My Dharma Talk with ChatGPT

I spent many hours over the past Holiday season talking to ChatGPT. After one session, I was so impressed with its ease of use and answers that it was clear A.I. did reach a new level. It’s certainly not “THE SINGULARITY”, but it’s now good enough – even after the current over-hype eventually fades – to drive real change in the way we interact with computers.

I’ve mentioned some of my Holiday fun with ChatGPT in a prior Christmas 2022 blog. So apologies for some repeat.

By “good enough”, I mean that in a business domain – which is a narrower scope than the domain of everything – something like ChatGPT could be effective enough to substantially smooth out communication friction between people and software applications.

Currently, people are restricted to a small set of things they can query from a software application through whatever a human programmed for them. Queries need to be unambiguous using a strict procedure or syntax (like SQL). But the real world is full of ambiguity that manifests as a great deal of friction.

I decided to shift my career all-in on A.I. and placed my bet on an A.I. platform. I thought that Azure’s Cognitive Services made the most sense, since ChatGPT is deployed on Azure, my background is mostly with Microsoft products, and I’ve been playing with Cognitive Services for a bit already. So I immediately began prepping for Microsoft’s Azure A.I. Engineer (not A.I. inventor or researcher) certification – which I did just pass after five weeks of intense practice and study. It helped that I have been using the related Azure ML platform and have been an Azure Data Engineer for quite a while.

It’s not so much that I’m excited about the opportunities this now “good enough” level of A.I. will unleash on the world. In fact, I’d rather it didn’t exist. My keen interest is more along the lines of “keep your friends close and your ‘enemies’ closer.” As we all hold in our consciousness from many movies, all things A.I. can suddenly get out of hand. So it’s good to know the subject first-hand by shifting my career directly to implementing A.I. systems. Of course, in a way that we’re all cognizant of unintended consequences (a “responsible way” as Microsoft and Open AI say).

The A.I. phenomenon of ChatGPT isn’t really all that new. We’ve been living with Siri, Alexa, and bots for years. Even ChatGPT’s prior incarnations have been around for a few years. But the results hadn’t been much better than a common Google search – which is usually what I ended up getting from Siri and Alexa anyway. What struck me is that ChatGPT seems good enough to usher in a profoundly new common interface between human intelligence and machine intelligence.

Encountering ChatGPT reminds me of when I first played with a GUI (graphical user interface) around 1990. It was Windows 3.0 and it sure was a profound change over the 24×80 matrix of ASCII characters of the interfaces I had been building for a decade. Yes, pictures (icons) are worth a thousand words.

GUIs raised the ease-of-use to a point that opened the door to multitudes more. It removed massive friction between human thought and machines. Cell phones opened the door to even more folks, but I think of it as more like GUI 2.0.

Following is part of a little dharma talk I had with chatGPT. I thought of asking it about Buddhism since it kept telling me: “As an artificial intelligence, I do not have the capacity to wish or desire anything …” That sounds kind of Buddhist to me … hahaha.

My first question is a simple recall question. “You never step in the same river twice” is often attributed to the Buddha. I’ve learned that it is from the lesser known Heraclitus, but it is something that the Buddha must have said every day in many different ways. Who knows what either actually said since the quote is a translation into a language that didn’t exist back in their day.

I think “You never step in the same river twice” being attributed to the Buddha is due to Herman Hesse’s classic novel, “Siddhartha”. The impermanence of a river is kind of the punchline of the novel.

I’m impressed that ChatGPT didn’t strongly assert (ChatGPT often comes across as very “opinionated”) that it is a quote from Heraclitus, dialing it down to “… often quoted with saying …”

This follow-up is more impressive because ChatGPT needed to abstractly consider the work of Buddha:

I would have been more impressed if ChatGPT answered in terms of impermanence as central to the Buddha’s teaching, not just one of the subjects. (Suffering resulting from clinging to what is impermanent.) And my question was a little leading. It would have been more interesting to ask “Which philosophers might agree?”

This second follow-up is indeed impressive:

For one, it did keep track of our conversation. That ability is a big part of the secret sauce to ChatGPT. My question wasn’t a leading one either. for example, “Are the river banks part of the metaphor of impermanence as well?”

In a later conversation, I did ask ChatGPT which philosophers would agree with “you never step into the same river twice”. It again mentioned that it’s often attributed to Heraclitus but listed (along with brief reasonings) Buddha, Lao Tzu, Stoics, Heraclitus, Jainism, Hinduism.

Not on the list was Christianity. So I asked it and received this remarkable response:

Let’s look at one of ChatGPT’s recommended verses:

“Why, you do not even know what will happen tomorrow. What is your life? You are a mist that appears for a little while and then vanishes.” – John 4:14

That sure does sound like impermanence and being in the present!

As I mentioned above, ChatGPT isn’t the A.I. of the so-called “singularity”. It’s intelligent unlike any human in that it is at least “familiar” with much more than any single human has ever readily known. However, its inferences are usually commonplace. Meaning, I could have readily found that advice on Google for the past ten plus years.

It’s at times fascinating, but mostly in the ways that the thoughts of still free-thinking children are fascinating – very intriguing, but usually not helpful towards the immediate problem of our daily lives. At times, ChatGPT’s responses are somewhat impressive as in the case of this little dharma chat.

To be clear, ChatGPT is far from a deep expert at anything. As long as my questioning is wide of breadth but shallow depth, it’s much more “knowledgeable” than anyone I know. But soon after the depth of my questioning deepens, it starts to show its faults. I think there will be another modest “A.I. winter” (a San Diego kind of winter) soon after the over-hype phase settles down. But as the Dot-com crash in 2000 adjusted what was then over-hype, the Internet eventually lived up to its hype of the Dot-com era.

On the other hand, there exists a real chance that the level of A.I. could improve very rapidly. I get this feeling that the impressive quality of ChatGPT is (or was) somewhat surprising to its makers. Even if they can explain why now, that’s with some level of the benefit of hindsight.

With all that said, I’ve been using ChatGPT rather heavily during the course of my daily work over the past few weeks. Every day during the course of my work I encounter subtle differences between concepts – the sort of subtle difference that somehow makes a lot of difference in practice. These are the sort of problems I might have spent hours researching. But ChatGPT can often provides the punchline of a direction from which I can work my way backwards into the details. So it is valuable to me well beyond Google, and I cherish these days that it’s free.

I also predict that rather than pushback on A.I. winning out in the short term, there will be a notion of “private A.I.s”. That would be A.I. models trained on private enterprise data. There’s much that can go wrong with everyone working off the same data. It’s no different than reaping the value of any other variation in something.

Additionally, there could be an inherent problem with reinforcement learning performed by human selection. Meaning, people providing ChatGPT feedback on what is a “good” answer. This is no different than the results of a survey involving an inadvertently biased sample.

Lastly, I was amused to see what a terrible evangelist I am for the Eternal Fishnu:

John the Baptist or The Bab, I am not … hahaha. Fortunately, the size of the Eternal Fishnu’s Sangha isn’t one of my KPIs.

Faith and Patience,

Reverened Dukkha Hanamoku