Recently, a friend asked me about my position towards the development of AI. He was amazed by the development of the technology, particularly its ability to generate videos.
This story is my attempt to re-formulate my reply to him, plus expand on the topic.
Disclaimer#
First, I feel I have to make this disclaimer.
I may ask some AI to help me formulate a few words, but all of the texts are written by me. I have a great deal of experience as a writer, it’s just natural to me, to write.
Still, having this option to ask a robot some natural-language request is of a tremendous help. I use AI to ask for some word. Sometimes I want to use a very specific word, which I have difficulties to recall. I have no idea how to search for it, yet when I ask AI, it finds the word I’m thinking about pretty quickly.
First Interactions#
When my foster dad was alive (he was a software engineer), he tried to explain to me what Artificial Intelligence is. It was decades before it became popular, and I was still a teenager. Not meaning I have other things in my head (which was also the case), but rather I felt like the whole matter is too complicated for me to grasp. All I remember, is that I’ve got that impression that it’s mostly math. I’ve been very good at math, top of my school, yet, I’ve struggled to grasp this then.
The Hype#
I’ve jumped this current AI hype train too late, it was about a year ago, late 2023. I was super-busy, natually having no resources even to play with some famous websites, like ChatGPT.
Had three (four?) gigs at the time, including kids, work, rostymo gig and some remains of my personal life.
First thing that I tried was ChatGPT by OpenAI.
My first ateempts to use the AI were primitive. I tried to ask it some questions I’d ask Google Assistant (or Siri before that).
You may want to explore my previous stories on:
I started asking questions about the weather in Hong Kong, the time zone of San-Francisco and alike. I believe LLMs are destined to fail at that, because they’re ought to be double-checked on important data. Nothing beats Google Assistant for me to this day.
Texts#
Then, I tried the tool for writing some texts. This is where the tool shines. For me, it showed itself great at re-formulating things, formulating technical texts (e.g. documentation or some web-related content for SEO).
Over time, I learned that the tool is good, but you have to double-check it.
I was testing it translating things, and it was much better than, say, Google Translate. Google Translate is still awful at translating sentences, it yells ‘machine translate!’ I wonder why, isn’t Google AI-first company since many years ago?
Still, Google Translate is awful for me, I use it only when I want to find some word, and I forgot it, but I remember it in another language. When I need to understand some word, if it’s Enlgish, I use OED, Oxford English Dictionary. If it’s other language, I may translate the word into English, and use OED if I don’t know the word.
I tried to use it as a helping hand with my writings with this website, but I abandoned this idea after a couple of tries. Even when it helped me to come up with some structure, it forced me to follow this arificial structure that bears no logic to me. That worsens the whole story-telling experience for me.
As I’m a natural-born writer, I see LLMs as of a great help to improve my writings, but not replace them entirely.
Images#
With the images, situation is somewhat more complex. I’m not an illustrator. Not a true one. I like to draw, and I take it as a hobby, but not a profession. But my wife is. She’s an illustrator, she does exactly that her entire life. Many of my previous partners were illustrators too.
I don’t want to be a luddite here, but I see the issue illustrators see in the development of AI. It can easily replace them. Having these models trained on some illustrations, these days we can generate images easily and within seconds.
That’s a problem. And I have no solution to this problem. Neither moral, nor technological. I don’t know what to do. I can demonstrate you what’s up with the case of one project.
A client of mine, they have this annual project, to make some postcards to send across their hundreds (or even thousands) of employees. It has to be some generic funny ‘Merry Christmas and a Happy New Year’ postcard. Nothing too fancy, nothing too complicated. Yet, I’ve been failing this project for years!
If you’re interested, you may want to explore this as its separate story.
Making images with AI is easy these days. However, we need to find out what’s etical for it, and what’s not. That’s for sure. Yet, the technology is very impressive.
Videos#
This year, I’d decided to expand on the old horisons and go with the video. This time, we used Synthesia software, synthesia.io. It’s truly impressive. I may want to write a separate big story on Synthesia, it’s beyond good.
Fast forward, we created this postcard with speaking Santa. We created many of them, for different occasions and in different languages, each one of them took about 5 minutes to create. Or, to place it differently, 1 minute to create and 4 minutes to render in background.
Look at it, it looks super natural to me.
Again, it’s not the best possible work, but it’s more a demonstration of what is possible, within minutes.
Software#
I’ve been using ChatGPT to craft custom automation scripts for my Arch Linux system. They’re mostly in bash.
I plan to open-source them as soon, as I’d clean-up the code, it’s a bit of a mess now. I’d like to test my scripts over time, and see what works and what isn’t that good as I think it is.
I’m not good in bash, I’m able to read the code and understand all of it, but I’m unable to write that code (at least that easily and quickly) myself. I succeded with most of my scripts. ChatGPT-4o writes them very well. While I found ChatGPT-4o mini mostly inferior.
Too Complicated#
The only one time ChatGPT failed me spectacularly, is when I wanted to make a conversion of my notes. They were imported from Evernote to Notion, and then exported. I chose this weird way, because Evernote become so shitty, it couldn’t export my notes on its own. Only one-by-one. Possibly, it still able to export them with their Windows software, but I have no Windows and not really willing to seek it for the task. So I tried to import them into my Notion account.
I wanted them to be converted for my personal notes websites, which is a separate task on its own. I tried to make it into one super-complicated bash-script with hundreds LoCs. The more complicated the whole thing was becoming, the worse the AI made it after each iteration.
That’s a very wild use-case, and I believe that could be a separate task for the Notion team, so no wonders I failed to do that within a couple of hours. Yet, I succedeed to half-ass it. My bet, I could sew a few simpler scripts to achieve the result I want. I abandoned the task, postponing it till some more free days, as its of a very low priority. I just expected to solve this task within one spare morning.
Co-pilot#
However, generally, I’m very happy with this AI help. For me, the primary benefit of having this bot is to have someone always available to chat over the things. I can ask human-language questions, and it can answer. Upon formulating my task, I understand it better myself. Without this virtual assistant presence, there’s no real intent to formulate the whole thing in the first place. To me, it works similarly to having a blog. I have some super unpopular blogs, yet it’s still much more enjoyable to write there, than to write it for myself only. To me, it has some quantum vibe of this observer effect.
When there’s no observer, I have no motivation to write. When there’s even minor activity (no need for millions of views), it just becomes different. The same way with having this co-pilot.
Worries#
I think an AI assistant can help re-write (refactor) the code quickly and easily. However, I’m worried about it having some hidden cost. E.g. it may break other things in upredicted ways. So, again, I’d say this AI help should be used very carefully and by a competent party only.
If we’d take the refactoring case, the person refactoring should understand the code very deeply (more likely that should be a maintainer, or the creator themselves). And that party should know the programming language involved on a very deep level, so to not be tricked by the bot that says gibberish in a very confident way. It ought to be constantly double-checked upon.
I believe the development of AI would influence the future of software development in a very rapid way. And my concern is, it won’t be positive only. There would be cases of very unoptimised and mediocre software. Even if created quickly, it can bear a problem: it fills some niche, and the better-written alternative won’t be developed out of not necessity.
It may help for rapid development of some open-source tools, for sure. Possibly, Gimp 3 would eventually be released, and the next time they port from GTK-3 to GTK-4, possibly it would take them less than 12 years. So in general, I believe it’s a good thing.
Open-Source AI#
There’s this development of meta, called Ollama, ollama.com. You can run it locally. When I have a few machines able of running most of their models, I had no chance to test them, unfortunately. I guess, I’ll try some of them the following year, so I may write another story about this thing only. I believe it’s worth the attention. And I believe that to be the future of the AI development. The local open-source models.
Conclusion#
At the moment, I have no scepticism towards this technology. It revolutionises how the computer work is done. It may take work from us, but as any other technology, it opens us a way of creating something more complex and thoughtful. It takes the rountine off our shoulders, by allowing us to concentrate on something more complex to be created within the same time-frame. Allowing for much more impressive results. It cheapens many things, so we may get to see more impressive special effects not only from Hollywood vendors. It should help cartoon creators to maybe focus on something more meaningful. It should help software engineers to test some prototypes very quickly. It should revolutionise (no less) our medicine.
On the other hand, it’ll fill many niches with not worthy of our attention products. E.g. my website, being carefully crafted manually, could be generated by AI entirely. Would one want to read and explore it? Yeah, probably the first time as the demonstration of the technology. But in general?
It’ll pave the way for bad actors, e.g. ruzzian troll factory are already much more advanced than they were mere years ago.
As any modern technology, it just changes what things are.
If we, humans, could live much longer, we’d probably noticed things are at constant change. That’s a natural way of the universe.
P.S. Cultural Impact#
Also, I forgot to mention the already-happenned cultural impact. Many of you are likely seen all these AI-generated postcards and pictures.
- Those that I’ve seen are mostly religious, e.g. with Christ walking and talking. Apparently, they work very well on the religious audience.
- Also, I’ve seen some manipulative things, e.g. ‘they serve in Ukrainian army and nobody wished them happy birthday’ innocent things made by some popular social-media channels
Now, we’re entering the personal AI fun-time, when one can generate a funny personalised videos, targeted at one person, to be watched once. (Or several times, before it’ll be forgotten.)
Ecological Impact#
I’ve seen discussions following this topic. Usually they’re from the priviledged societies. Those who can casually travel with planes for no particular reason, multiple times a week.
Considering the context, I see this as mostly irrelevant. There are many other places we can cut out ecological impact, without even extreme measures.
Reply by Email