• The development of any software program, including, but not limited to, training a machine learning or artificial intelligence (AI) system, is prohibited using the contents and materials on this website.

The A.I. Revolution (ChatGPT, Stable Diffusion, Bard, MidJourney, etc)

GRtak said:

ChatGPT-maker OpenAI fires CEO Sam Altman for lack of candor with company​

Click to expand...
Well, that was quick.

venturebeat.com

Sam Altman will return as OpenAI CEO, New board announced in a dramatic turnaround

OpenAI announces major organizational shift, with Sam Altman's return as CEO, the introduction of a new board and substantial reset for the AI research lab.
venturebeat.com venturebeat.com

Sam Altman will return as OpenAI CEO, New board announced in a dramatic turnaround
 
Tech company shenanigans
 
WTF🔥

Everyone who downloaded this data has CSAM on their harddrive 🔥

Largest Dataset Powering AI Images Removed After Discovery of Child Sexual Abuse Material​

Click to expand...
www.404media.co

Largest Dataset Powering AI Images Removed After Discovery of Child Sexual Abuse Material

The model is a massive part of the AI-ecosystem, used by Stable Diffusion and other major generative AI products. The removal follows discoveries made by Stanford researchers, who found thousands instances of suspected child sexual abuse material in the dataset.
www.404media.co www.404media.co

Dumpster-fire-pic-3942990763.jpg
 
For the fun of it, and because somebody didn't restrict access to it on our work 365 accounts, I have started playing around with MS Copilot, which incorporates chatGPT4.0, DALL-E and other shenanigans. And honestly, for those of us who mainly do "thought work", which often enough involves combing through information and more often than not just needing a basic understanding what certain things are about - be it tools, laws, APIs and suchlike, this tool is frankly insane. Stuff that would've normally taken me a full working day I can get done in an hour or less - if it lends itself to it. As far as I can tell so far, the current copilot doesn't actively lie to me or make shit up - as long as I clearly reference a law, API or other clearly defined text that already existed in 2021 somewhere.

... so I will now spend the efficiency gains of this on random bullshit, as per use :D
 
Here's a fun use for AI, that's effectively harmless but shows the potential harm it can do for creatives.


View: https://www.youtube.com/watch?v=NWwLm2pv7QI

I found this because I have (somewhere) an old Brother AX-310 electric typewriter just like this, and I want to build an interface to allow me to use it as a daisywheel printer. Purely to print authentic copies of Red Dwarf scripts. :LOL:
 
Last edited:
Found one in the wild that I find odd.

A friend on Facebook shared the image below.

1000015130.jpg


It just didn't look right. There's AI artifacts that I just couldn't understand. Plus, I figured this was an old photo. Taco Bell sign, 7/11, cars, etc. all look 80s/90s.


1000015131.jpg


A quick Google search found the above image, on Pinterest of all places. While it's grainy, you can see what the small sign on the 7/11 says now and things make more sense.

Why would someone bother to "clean" the original image?
 
CraigB said:
.

Why would someone bother to "clean" the original image?
Click to expand...

I've come across this a lot in video: people applying AI upscale to low-res videos. It can work OK if the jump up in resolution isn't too big, but turning something like 240i into 4K is too big of a jump and it looks worse, like a scene from Waking Life.

People seem to respond to, and are impressed by, the "new" smoothness of the old content but don't realize or don't care that they are missing out on detail you could still resolve in the lower resolution video.

I suspect it's the same people who, when they get a new phone, make sure their camera has any smoothing or beauty filters turned on and cranked to 10x, and every selfie looks like the lense was smeared with vaseline.
 
Last edited:
I have the Topaz AI tools and have experimented a lot with upscaling of photos and videos. I find it really interesting. I wouldn't be happy with the result for that image, and I'm often not happy with the results of upscaling low quality videos, particularly VHS where it's quite grainy. I wanted to re-watch the S-Class cottage part of TG the other day so threw my 480p copy through the Topaz enhancer to see what would happen. The result had its merits but was awful overall. My photo enhance tool did a similar job with that original Taco Bell image, it's way too grainy for it to do a decent job.

There are some good uses, the video tools can also produce intermediate frames to increase framerate, produce slow motion clips or eliminate interlacing. I'm (very slowly because it takes hours) going through my Blu-ray copies of Red Dwarf and de-interlacing the video from 25fps to 50fps. It doesn't upscale the video so there are no nasty effects and the result looks pretty damn good. I might be able to upload a sample on my second YouTube channel.

For upscale photo and video tools work better with CG images or videos. So AI created images will upscale well, I've played with this too. CG artwork like that produced by Jim'll Paint It upscales well. Old video game concept art upscales well, although it's funny that where the edges of any features will be upscaled well, any textures will still be blurry. I'm not sure if that's good or bad.

The best result I've ever seen with AI video upscaling was with my old Forza Motorsport 4 replay videos. These were uploaded to Microsoft's servers directly from the Xbox 360 and were downloaded in 720p 30fps WMV, decent at the time but far from impressive now. I scaled these up to 4K 60fps using Topaz and frankly the results blew me away. Here's a compilation I made many years ago using the source 720p footage and below is the more recent re-make, which was made from scratch using upscale footage rather than just upscaling the output of the original. You can see where some of the road surface textures remain soft but the detail in the cars is amazing.

720p:

View: https://www.youtube.com/watch?v=Hj3CWcxdK3Y
4K:

View: https://www.youtube.com/watch?v=5obv_3jwrFU
If you don't fancy comparing yourself here is a snip of two frames. The 720p frame was resized in Photoshop to 3840x2160 for the comparison, using bilinear re-sampling. Years ago that would've been considered as good as we can get.

Resized
Cortina_Resized.jpg


AI Upscaled
Cortina_Upscaled.jpg


You can see it messed up the Lotus logo, but otherwise it has pulled a lot of detail seemingly from nowhere, particularly in the reflections on the chrome.

Weirdly this improvement isn't seen when running 1080p captures from the game through the same AI tool, not sure what that's about. Those are better resolution/framerate to begin with but don't look as good as these upscaled clips.
 
Last edited:
Car customers, journalists, drivers and literally everyone else in the known universe except bean counters and marketing departments: "Can we please have knobs and switches back?"

Volkswagen marketing department: "ChatGPT seems to be all the rage right now. Everyone stop what you're doing!"


View: https://www.youtube.com/watch?v=S--NKBa6sfQ
 
Perc said:
Volkswagen marketing department: "ChatGPT seems to be all the rage right now. Everyone stop what you're doing!"
Click to expand...
What I find particularly infuriating about this kind of thing: everybody seems to equate AI to ChatGPT now. Even people who should know better. Nobody seems to understand it even just a little. Just running along the hype train, rationally thought or any actually sensible use case for AI or ML be damned we just want to chat with the car. Morons.
 
eizbaer said:
What I find particularly infuriating about this kind of thing: everybody seems to equate AI to ChatGPT now. Even people who should know better. Nobody seems to understand it even just a little. Just running along the hype train, rationally thought or any actually sensible use case for AI or ML be damned we just want to chat with the car. Morons.
Click to expand...

And we've had voice synth for literally ages, but all of a sudden it's "AI".
 
Let's have a new ML generated songs about the Dunsfold Aerodrome.. in a few different styles, but using the same lyrics for each; What's impressive is the lyrics came together pretty nicely with such a small prompt and time window, and no major flaws with the vocals or backing audio. Guitars get muddy and the vocals occasionally get pitch shifted around, but for throwing together a music idea in 30 seconds?

Power Rock Ballad?


Anthemic Hard Rock? (audio levels go in and out)


Big Band?


Opera? (Warning: This gets loud.)
 
  • Like
Reactions: lip
You must log in or register to reply here.
Top