is AI gonna take our jobs?

If you are a spelling bee pronouncer. :)
yeah that's what I thought. but it does prove the point that these programs can't be trusted to reliably solve basic problems. and if you get into more complicated logic puzzles, they're a total wreck but very confident in their responses.

more on point, coke came out with an AI-generated ad recently. My basic impression was that if you weren't paying close attention, you wouldn't notice that it was AI. but the fact of the matter is that the human beings, the dog, and the close-up of Santa's hand all looked weird and lifeless. it would have been so easy to shoot those elements with real-life subjects and place them in the AI backgrounds. that's been my prediction for a while -- that AI replaces CGI and greenscreen stuff but that the subjects will still be filmed (and probably on a soundstage with good lighting) just because the result will be better and the cost is marginal for any high-value production.

Then you still have technicians working with AI instead of working with CGI. it will probably take them less time so it may be a smaller invoice. but the basic contours of the production will probably be similar to pre-AI.

Also, shouldn't you start disclosing your holdings on your AI posts? ;)
 
Trust is a huge part of everything. The tech folks are very prone to looking past faults just to get their products out there. Its entertaining to make goofy pictures and generate content but when it comes to large money, serious stuff, I do not think it is close to being there and the 'makers' will never admit that. So for me it is more about trusting the makers than the technology potential.
 
Interesting thread on the state of AGI research. I'm linking to the post that shows an example of a logic problem that is easy for a human to a solve but still impossible for a state of the art AI using thousands of dollars of compute time. But there are numerous posts in the thread prior to this one


As an aside, my take on using Sora for several dozen generations - equally fun and frustrating. It's pretty obvious that what it does well is what most closely matches its training data. So basically it's a copyright-laundering machine. But it has lots of potential. I think stock footage and animators will be most impacted by its current capabilities.

Although I expect it to improve somewhat, if it is limited by what is available in its training data and if the model has already exhausted the high quality training data available, then I don't expect it to qualitatively improve
 
Although I expect it to improve somewhat, if it is limited by what is available in its training data and if the model has already exhausted the high quality training data available, then I don't expect it to qualitatively improve
I think this is an area where the potential of AI will fall short. The internet is largely seen as a limitless venture. AI carries that same perspective from people now. But as more folks restrict their IP from being used by AI, it might just come up short. Not what the AI creators are seeing but it seems their plans are based upon people giving then free reign to everything. The reality will probably fall into the middle. "Good AI" will end up costing a lot which might make the whole thing cost prohibitive.
 
I think this is an area where the potential of AI will fall short. The internet is largely seen as a limitless venture. AI carries that same perspective from people now. But as more folks restrict their IP from being used by AI, it might just come up short. Not what the AI creators are seeing but it seems their plans are based upon people giving then free reign to everything. The reality will probably fall into the middle. "Good AI" will end up costing a lot which might make the whole thing cost prohibitive.
I think I agree with you. And I'm guessing that most of the high quality available video sources have already been mined, such as YouTube. It may be impossible to reap new high quality video at scale going forward because people will start using AI in the production of videos and AI can't train off of AI-created material. The model just turns to mush.

Here's another example of an epic fail at basic reasoning that someone just posted on Twitter. It took chatgpt over a minute to analyze this

1000028375.png
 
Okay all those meticulous jobs that are in post production are gonna be replaced by AI. Given the advancements of technology, it's going to be harder to find a fit in the film industry given pre-production and post-production. Production people are gonna be the most salvable and in many ways a bit of pre-production, but I can imagine like literally half of the film industry disappearing from this technology. Marvel is going to replace a ton of its post production office with this technology and the advancement of it's own. What's going to happen is the next big technology boom in the industry ever since Jurassic Park.

Now for me to get a footing in filmmaking post-production is going to be out of the picture as is pre-production, which leave production, but production I don't get along well with people, so I'm kinda screwed.

Even production it's kinda losing opportunities for me.
 
Last edited:
I think I agree with you. And I'm guessing that most of the high quality available video sources have already been mined, such as YouTube. It may be impossible to reap new high quality video at scale going forward because people will start using AI in the production of videos and AI can't train off of AI-created material. The model just turns to mush.

Here's another example of an epic fail at basic reasoning that someone just posted on Twitter. It took chatgpt over a minute to analyze this

View attachment 5712435
Is this semantics at play or something? The what is the definition of why red and blue are the same length?
 
Is this semantics at play or something? The what is the definition of why red and blue are the same length?
I don't know but I think the machine is being too clever by half. It knows that there is a paradox concerning apparent line length and it thinks you're more likely to ask about that?
 
Parents of the OpenAI whistleblower whose death was deemed a suicide claim that he was murdered:
"The tragedy in San Francisco happened three months after Balaji accused OpenAI of violating copyright law in their development of ChatGPT."

I've been saying it -- LLMs are a copyright-laundering machine.

By the way, a handful of other whistleblowers died in the US in 2024 under mysterious circumstances, including at least 2 who blew the whistle on Boeing.
 
I don't think it's a question of whether it's going to start taking jobs but more of a question of should you embrace it?

It's a new technology that's not going to go away and it's going to continue to advance

Kind of like stop motion days moving into CGI (bad analogy?)

I think we should embrace it because it's coming no matter what

I think in our time we're going to see a full feature length motion picture created entirely by AI and I bet you it won't even be that bad
 
They already have a few things going out there...not 40+ minutes but it's all the same; just add more minutes of computer generation:

 
They already have a few things going out there...not 40+ minutes but it's all the same; just add more minutes of computer generation:

I look forward to Netflix movies getting even worse.

At the moment, saying that movies are created by AI is a bit misleading. It takes a lot of work to prompt the machine to give you something close to what you want. Even then, there's a lot of post involved. So really what they're doing is cutting out principal photography. Which is huge if it comes to pass.

But I read the making of info on the balloon head Sora short. The guys who made it said that they basically generated a lot of footage they thought might be useful and then edited it like they would any other shoot, treating the generated clips like raw footage. They also said that there were some weird problems they couldn't surmount. Like for a particular shot, the program gave them the wrong color balloon no matter how many times they asked it to correct the balloon color. So they just changed the balloon color in color correction. But my assumption is that that shot was generated from an actual shot of a balloon with a different color and that may be why it kept giving them that other color.
 
Lol openAI is so screwed once the litigation really gets going

"OpenAI has long refused to say whether its Sora video generator was trained on YouTube content — but its propensity for generating videos that look a whole lot like real gaming streamers suggests it did.

When TechCrunch put Sora to the test, its reporters found not only that it could generate videos that were strikingly similar to real-life gameplay of "Super Mario Bros" and "Call of Duty," but also spat out what appeared very much to look like the streamers Raúl Álvarez "Auronplay" Genesand and Imane "Pokimane" Anys.

Though OpenAI claims it has guardrails on the way it depicts real people, it doesn't seem that reporters had any trouble getting it to spit out a video of Anys — though she did end up looking pretty monstrous, with the uncannily exaggerated features distinctive to AI depictions."

also, "OpenAI Admits It Needs Vastly More Money Than It Thought"
 
Back
Top