AI Rant 4: Things are changing too quickly
It is next to impossible to follow what the hell is going on in the AI world anymore. I don’t know if there is a huge rush to announce AI projects ASAP to try and get ahead of competition, or if things really are growing exponentially.
I made a post about AI Art’s dangerous effect on the 3D community, only to follow it 3 weeks later with a post about 3D artists potentially cooperating with AI using controlnet. That’s how fast it is moving.
Early this year I posted about potential issues regarding copyright issues with AI art programs like Midjourney and Stable Diffusion scraping the internet for images without compensating artists.
Suddenly Adobe come up with an AI art project of their own called Firefly which uses 100% licensed and public domain images with a promise to compensate artists who allow their art to be used. Problem solved? Well it might take some time for others to follow suit, but it is a step in the right direction.
I wrote about the possibility in my AI Art post part 2, and here we are in a mere 3 months and it is already here.
Fun with Nvidia Canvas
Meanwhile, I found a fun free AI art tool I can use in my own images called Nvidia Canvas which can create detailed outdoor landscapes I can use as backgrounds for my 3D art. I can create 2D backdrops or 3D panoramas for my characters using the images both as background and lighting models to make the renders look lit to match the scene.
The image on the left is a drawing I made using a mouse. The writing was added later just to show you what each color represents. The image on the right is what the program generated based on my rudimentary drawing. I then save the image on the right and import it into Poser both as a background and as a HDRI light texture and got this:
Then if I want, I can throw this image into SD+controlnet and then get something like this:
It could use some final touching up, but that is a quick demo in going from AI to 3D back to AI. This is just a simple example thrown together quickly (30 minutes?) as a demo. Spending more time on this could result in some fantastic images.
And the AI game changes again! Chatbots can be copied shrunk to their essentials and run on a PC.
The financial drive for AI on companies is easy to understand. Provide these online services, charge people to use them, ???, Profit!
But what if people had their own personal chatbot they could run locally on their own computer, no online cloud required? Well we may be headed this way thanks to research being done at Stanford.
A new program called “Stanford Alpaca” is undercutting the cost of AI chatbot software from millions of dollars, to hundreds. Maybe even cheaper soon. Running ChatGPT quality programs offline from your hard drive will likely be a thing before the end of the year.
This will pretty much kill the AI enabled services market just as it was getting started.
The big question is, what happens when millions of people run their own unrestricted AI bots all over the internet?
Bottom line, attempts to control it are not going to go very well.
More Scary thoughts
If the 2020 election was all about controlling social media to control the public discourse…
The 2024 election will be about AI generated talking points to attract the most voters.
This technology is moving so fast, it is hard to see what evils can be wrought. I followed the progress of the Internet from savior to destroyer documented in my Internet Dream series, and I am pretty confident AI is in for the same bumpy ride. It already makes deep fakes and revenge porn practically trivial to create. What other evils can it accomplish in the future?
There will no doubt be some good things too, but how this balances out is anyone’s guess.