The Core

Why We Are Here => Hardware & Technology => Topic started by: ergophobe on August 14, 2025, 09:51:32 PM

Title: If you can spot AI content today....
Post by: ergophobe on August 14, 2025, 09:51:32 PM
You won't be able to next year.

I was going to add this to the "I'm sick of hearing about" thread because I'm getting pretty tired of people saying things like, "It's so obvious when it's AI content because it's just crap."

What they really mean is that they are unable to discern 80% of AI content and the 20% they can discern is so bad that it really bothers them.

My guess is that the people who complain the loudest about this are the ones least likely to recognize AI content. It reminds me of the 1980s when I would be standing with a couple of gay friends and a fourth person and someone would go by and the fourth person would look at the person going by and say, "That guy is gay. I can always tell," while being oblivious to the fact that he was talking to two gay men.

Anyway...

Check out the crusty old sailor video that Veo generated. Is it perfect? Not yet. If you played this for someone 25 years ago, would they think for one second that it had been generated from a simple text prompt? I think not, but I think they would notice a few odd glitches in the matrix - cadence. Still to go from a simple text prompt to that is mind boggling.

https://deepmind.google/models/veo/


Attached:

1. Climber
A still photo that I generated using Gemini with a prompt like (I closed the window so I don't have the exact prompt): "Create a photorealistic image of a rock climber climbing a crack high above the Valley floor. Add 4-6 camming devices to his harness."

Note that it doesn't seem to know what camming devices are, so it doesn't add them.

On the one hand, I would say it still has very much an AI feel, a strong Uncanny Valley feel. If you showed someone this image 10 years ago, they would likely assume someone had gone overboard with the airbrush and softening tools in Photoshop.

The last time I did this, proportions were wrong, positions didn't make sense in terms of the physics and sometimes not even in terms of the geometry. In six months it has gone from obvious and clear fake to a very good fake with a slight uncanny valley feel.

And keep in mind this is a free tool with minimal prompt refinement. I'm guessing that the free tool has fewer diffusion steps, which is what gives it that vaguely fake, airbrushed feel. With more diffusion steps, you get more detail and more photo realism (to a point).

2. Singer on car
The second attachment is from the Midjourney subreddit, sorted to put the "best" posts close to the top. Some of them are still obviously AI-generated, but some are getting very hard to tell. I thought this image of a woman on the hood of a car would not have triggered any Uncanny Valley alarms five years ago.

My instinct when I look deeply at the second image is to say there is an error in rendering the face around the right eye, but now that I am called in to troubleshoot my wife's portraits sometimes, I realize that in fact this is probably correct.
Title: Re: If you can spot AI content today....
Post by: ergophobe on August 15, 2025, 07:25:26 PM
Quote from: rcjordan on August 15, 2025, 06:22:29 PMGood looking, but I had to filter her [Syndey Sweeney].

Not nearly as good looking as the AI models I have been seeing ;-)

Note: I know nothing about Sydney Sweeney, so I'm not talking about her in particular, but based on what I'm seeing, if your primary selling point is that you are beautiful, you will be out of a job in 1-2 years unless you are already famous enough that your name means something for an endorsement.

In particular, pay attention to products for older people. This may be because I'm now of an age where I qualify for Social Security and senior this and senior that, but I am seeing a lot of super buff AI models, men and women. Basically the bodies of 36yos on steroids, but with grey hair and wrinkles.

If you go down that rabbit hole for 6-8 clicks, you quickly go from seniors to "perfect" women. There are currently three ways to guess that something is AI

1. Artifacts - essentially the "airbrush" effect I mentioned in still photos or something resembling JPEG artifacts when looking at AI video. I figure that just like we rarely see this with JPEGs anymore because bandwidth and compute power make higher res and higher quality the standard, this will be coming to AI soon.

2. Too perfect. In the same way that for a while you could spot AI writing by the lack of typos and grammatical errors, these models look like "perfect" humans. I expect this to change soon. We've already seen it with text where students are adding imperfection prompts. I think AI graphic artists will be adding imperfection prompts to their images of humans too

3. Still lots of the world that LLMs don't understand. Climbing is pretty mainstream these days, but not like golf. So the AI images I've tried to create of climbers and the ones that I've seen online get the gear wrong. Sometimes it's a jumble of junk that at first glance looks like climbing gear, but upon inspections is a random amalgam of aluminum.

Still, based on what I'm seeing in #2, the images of those with "perfect jeans" like Sydney will not be human in the very near term.

Also, a thing I'm not quite understanding, is that there are some AI models that appear over and over and over. I know going back years in 3D modeling, models (meaning objects, not necessarily human models) would get shared and remixed. So you would see the same cars, same people, same houses in 3D video animations. I think there are some "stock" models that get remixed in the AI world now too. I think there might even be a couple of stock models on their way to becoming famous.
Title: Re: If you can spot AI content today....
Post by: rcjordan on August 16, 2025, 12:17:57 AM
IMO, the Old Sailor you posted is already good enough quality to displace human talent in maybe the majority of online ads.

Nice post, btw.
Title: Re: If you can spot AI content today....
Post by: rcjordan on August 16, 2025, 01:59:55 AM
+

add porn actors to the Dead Jobs Morgue
Title: Re: If you can spot AI content today....
Post by: littleman on August 17, 2025, 11:35:58 PM
>porn

Bodes poorly for future reproduction.  Reality is going to be shockingly imperfect to reproductive age humans pretty soon.  Stack that onto the pile of reasons birth rates are falling.
Title: Re: If you can spot AI content today....
Post by: ergophobe on August 18, 2025, 11:59:56 PM
Quote from: rcjordan on August 16, 2025, 12:17:57 AMIMO, the Old Sailor you posted is already good enough quality to displace human talent in maybe the majority of online ads.

Nice post, btw.

Thanks.

So... TikTok ads are already doing it at scale.
https://www.tiktok.com/@tiktoknewsroom/video/7382934469679959342

https://www.nytimes.com/2025/08/17/business/tiktok-ai-avatars.html

Google is supposedly offering it, but I haven't seen it. But they are doing real-time translation
https://www.youtube.com/watch?v=hyXqcsWOONo&t
Title: Re: If you can spot AI content today....
Post by: ergophobe on August 21, 2025, 09:38:16 PM
AI Isn't Coming for Hollywood. It's Already Arrived
https://longreads.com/2025/08/21/hollywood-stability-ai/

Mostly paywalled...
Title: Re: If you can spot AI content today....
Post by: ergophobe on August 28, 2025, 12:05:30 AM
4,000,000 AI stock video clips om Adobe site

https://stock.adobe.com/search/video?filters%5Bcontent_type%3Avideo%5D=1&filters%5Bgentech%5D=only&filters%5Breleases%3Ais_include%5D=1&filters%5Breleases%3Ais_exclude%5D=1&order=relevance&limit=100&search_type=filter-select&search_page=1&get_facets=1

In case wrapping breaks that URL, click here (https://stock.adobe.com/search/video?filters%5Bcontent_type%3Avideo%5D=1&filters%5Bgentech%5D=only&filters%5Breleases%3Ais_include%5D=1&filters%5Breleases%3Ais_exclude%5D=1&order=relevance&limit=100&search_type=filter-select&search_page=1&get_facets=1)
Title: Re: If you can spot AI content today....
Post by: ergophobe on September 07, 2025, 08:39:29 PM
AI models appear in Vogue for Guess
https://www.bbc.com/news/articles/cgeqe084nn4o

Don't worry, Guess exec says that AI will never displace real models. So that's settled. I think at this point, it's just a publicity stunt. I for one did not even know that Guess still existed as a brand, but I do now.

Here's a medley of AI models done with Veo 3, the same platform as the fisherman video
https://www.reddit.com/r/aivideo/comments/1lvl71v/vogue_ai_when_ai_models_go_full_savage_video/

Interview with the creators of the Guess ads
https://www.buzzfeed.com/natashajokic1/vogue-guess-ai-models-explained

For those who hate video, this one is 8 seconds
https://x.com/babaeizadeh/status/1924942128851124284

Here's the IG page of the people who did the Guess ads
https://www.instagram.com/seraphinnevallora/?hl=en
Title: Re: If you can spot AI content today....
Post by: rcjordan on September 07, 2025, 10:04:23 PM
>So that's settled.

Absolutely. hhh

did not even know that Guess still existed as a brand
Title: Re: If you can spot AI content today....
Post by: ergophobe on September 12, 2025, 07:52:04 PM
Quote from: rcjordan on September 07, 2025, 10:04:23 PMdid not even know that Guess still existed as a brand

Honestly, that's probably because you were happily married and moved onto other concerns by the late 1980s when Claudia Schiffer was the face (and body) of Guess and known to most young men with youthful hormone levels.

The picture on the left was iconic back then. That, I think, is the only reason I know about Guess jeans
https://www.elle.com/uk/fashion/news/g7385/claudia-goes-back-to-guess/

I think she was better known that Sydney Sweeney ;-)
Title: Re: If you can spot AI content today....
Post by: ergophobe on October 02, 2025, 08:29:59 PM
OpenAI released their new video engine. The results are stunning

https://sora.chatgpt.com/p/s_68dd5f4a0c9c8191a72bb485cb2f0a6b

Current selection of "front page" videos
https://sora.chatgpt.com/
Title: Re: If you can spot AI content today....
Post by: littleman on October 03, 2025, 03:09:20 PM
Very impressive.  I wonder how long it take for us to have full length movies that are 100% AI generated and people won't be able to tell the difference?

Also wondering if in the future people will gravitate more towards plays and other love performances to get away from AI.

I could also see people losing touch with what reality looks like.
Title: Re: If you can spot AI content today....
Post by: ergophobe on October 03, 2025, 07:29:57 PM
I've kind of been wondering some of the same things.

I've also been wondering when I watch a move today, how many are human actors with 100% green screen? The "sets" have become sumptuous because they aren't sets. I watched some of Untamed which takes place in Yosemite and a lot of the scenery shots are fabricated in some way or another. In other words, you see an element that exists in Yosemite (random shot of the south entrance) combined into a scene that definitely does not exist in Yosemite (random mountain from somewhere else or from AI).

I read an interview with an actor recently who said he really enjoyed making some movie because the director believes he gets better performances with live sets and so, unlike many of the actor's recent projects, it wasn't all done on a sound stage with green screen.

And that has raised your last question. People know the show is not actually filmed in Yosemite, but will the expect that towering mountain above the south entrance? Will they expect that the scenery surrounding El Cap on the show is what they will see when they see El Cap?

>> plays

I think of this with reference to what happened after the Milli Vanilli scandal. After that, MTV launched the whole Unplugged series where they basically guaranteed that, other than amplification, the musicians on stage were playing the music entirely by themselves and live.

And I think that's when concert tickets started to climb out of control. In an era when most recorded music in history is free or close to free, concert tickets are more expensive than ever, because people want to be there live. The recorded experience has less and less value even though, generally speaking, the audio quality is much better. The live experience has more and more value partly because it's "real" and partly because it's exclusive. Mostly because it's exclusive.

That said, people also consume more recorded music than ever and certain consume a lot less live music than they did 150 years ago.
Title: Re: If you can spot AI content today....
Post by: nffc on October 06, 2025, 04:55:54 AM
>Also wondering if in the future people will gravitate more towards plays and other love performances to get away from AI.

I think the "cool kids" will be hanging around in bookstores/live events etc. The question is will that be profitable for the venues? I've been putting on a few concerts over in Vietnam (pics on request) still hugely popular. I just think for "real" experiences the bar has been raised, everything needs to be extra special.

Title: Re: If you can spot AI content today....
Post by: Brad on October 06, 2025, 07:19:33 AM
Bloomberg Market Minute podcast (on Alexa) yesterday said many streaming companies were moving into live events.  Some of these at smaller venues like dinner theater stuff.  They believe people are looking for the whole experience when they go out. 

It remains to be seen if this really catches on, but if it does it could be a good time for theater actors. 
Title: Re: If you can spot AI content today....
Post by: littleman on October 08, 2025, 08:17:33 PM
>pics on request

Yes, please.
Title: Re: If you can spot AI content today....
Post by: ergophobe on October 23, 2025, 01:19:43 AM
Sora video of corporal talking to troops
https://youtube.com/shorts/4-2lkaQyq1Y?si=et3yuItQWtUfusp8

I keep wondering if the next generation of deep fakes will be videos using real actors and scenery and saying they are AI
Title: Re: If you can spot AI content today....
Post by: ergophobe on January 02, 2026, 12:44:13 AM
https://youtube.com/shorts/JaFiUNmZoGM?si=jzwKqLkeR6O3yiSc

Note how the bear and the moose merge into a bear-moose at the end. Still, the start is pretty convincing
Title: Re: If you can spot AI content today....
Post by: rcjordan on January 02, 2026, 01:42:09 AM
The anti-ai creatives on bsky are howling at the moon constantly. So much so that I'm filtering.  This is not a future existential threat for the various arts. It's here and its already entrenched. No putting this one back in the bottle.

>arts

AI 3d models are showing up quite often.
Title: Re: If you can spot AI content today....
Post by: ergophobe on January 02, 2026, 04:50:54 PM
> creatives

If you are a "code is poetry" type and think of coding as creative (which I do) Simon Willison has some interesting pull quotes

"[Claude Code] has the potential to transform all of tech. I also think we're going to see a real split in the tech industry (and everywhere code is written) between people who are outcome-driven and are excited to get to the part where they can test their work with users faster, and people who are process-driven and get their meaning from the engineering itself and are upset about having that taken away."
— Ben Werdmuller
https://simonwillison.net/2026/Jan/2/ben-werdmuller/

"[...] The puzzle is still there. What's gone is the labor. I never enjoyed hitting keys, writing minimal repro cases with little insight, digging through debug logs, or trying to decipher some obscure AWS IAM permission error. That work wasn't the puzzle for me. It was just friction, laborious and frustrating. The thinking remains; the hitting of the keys and the frustrating is what's been removed."
— Armin Ronacher
https://simonwillison.net/2025/Dec/30/armin-ronacher/

QuoteIn essence a language model changes you from a programmer who writes lines of code, to a programmer that manages the context the model has access to, prunes irrelevant things, adds useful material to context, and writes detailed specifications. If that doesn't sound fun to you, you won't enjoy it.

Think about it as if it is a junior developer that has read every textbook in the world but has 0 practical experience with your specific codebase, and is prone to forgetting anything but the most recent hour of things you've told it. What do you want to tell that intern to help them progress?

Eg you might put sticky notes on their desk to remind them of where your style guide lives, what the API documentation is for the APIs you use, some checklists of what is done and what is left to do, etc.

But the intern gets confused easily if it keeps accumulating sticky notes and there are now 100 sticky notes, so you have to periodically clear out irrelevant stickies and replace them with new stickies.
— Liz Fong-Jones, thread on Bluesky
https://simonwillison.net/2025/Dec/30/liz-fong-jones/

Et cetera
Title: Re: If you can spot AI content today....
Post by: rcjordan on January 02, 2026, 05:27:07 PM
Coding is a PITA.  Laying out the path to the end result is the art.

My bsky profile says "All my programming languages are dead" (to me). That's true.  But I still retain some knowledge of programming logic and some rough bits regarding html.  Combine the above with GPT-5 and a couple of million 1min fairy dollars and we can turn out some pretty awesome TM scripts.
Title: Re: If you can spot AI content today....
Post by: ergophobe on February 11, 2026, 10:45:23 PM
New test...

Prompt - just typed it in. No tweaking or revising. First and only try.

QuoteCreate a photographic image of a woman leading a crack climb in Yosemite. The figure should occupy about one third of the image, with Yosemite Valley and cliffs in the background. She should have at least one piece of protection in the crack and a full rack of quickdraws and cams.

The result is basically perfect (full-size versions in the footer - I just put the thumbnails here so there's no confusion... not that there could be).
gemini-climber-11-02-2026.png


Compare that to the previous effort at a "Yosemite" photo from back in August. I don't have the prompts, but this image took lots of prompt tweaking and about 10 tries to get this.
Generated Image August 28, 2025 - 3_03PM.jpeg

The differences are striking.

 - For starters, the image quality is vastly better. The rock texture is basically perfect.

 - it's the right gear. She has a full rack of cams that are clearly Black Diamond C4s, the most popular cam.

 - she is protected with a cam in the crack. One small quibble, nobody would hold a cam like that except perhaps a total beginner. The previous version clearly didn't "understand" crack climbing - it was using sport climbing bolts next to a crack. That kinda behavior can get you shot around these parts.

 - the body position feels right. As a general rule, in past attempts there was always something off with the position or the balance.  In the August image, the foot in the crack is just sort of pasted in there in an unlikely way and the foot on the face on tip toe also doesn't really make sense. In the new image, the feet are perfect and it looks like a climber trying to find a balanced resting position from which to place gear. Totally believable.

 - and that is most definitely Yosemite. So much so that I feel like I should be able to figure out where that climb is. The older image does its best to create something that vaguely feels like Yosemite, but doesn't quite know where it is.

 - the woman is completely believable rather than... how shall I put this? When non-climbers create ad campaigns for climbing, they typically cast super attractive people who usually don't look like they know how to climb. Obviously, there are some super attractive people who are amazing climbers, but this woman seems like the sort of person you would be *likely* to find at a crag... Or put another way, I would say she is quite attractive, but she's presented as a climber, not as a model on a climbing photoshoot. At least that's how it feels to me.

It is stunning.
Title: Re: If you can spot AI content today....
Post by: ergophobe on February 11, 2026, 11:05:19 PM
Okay... spoke too soon.

I asked for the following refinements
Quotemake the climb a little steeper, do it in portrait mode and make the rope come from below her, not to the side, with at least one piece of protection in the crack just below her feet

And got this.

Note that Gem changed the aspect ration just fine, but failed to move the rope as requested. The result that her protection is just hanging in space below her feet. The cams are the part that go in the crack. What's holding the camming device there is a mystery. On the plus side, it is noticeably the same climb with the same figure in the same place. In the past, I would lose all context and the climb and the model would change with every iteration.

So, big bump up, but still not ready to replace human models and photographers.

BTW, I choose climbing because unless they hire actual climbers and climbing rigger and climbing photographers, the images used in ad campaigns and movies are almost always ridiculous. So I figure it's a good test in that there are millions of climbing images out there now, but it's also not like basketball or golf that are utterly mainstream.

 
Title: Re: If you can spot AI content today....
Post by: rcjordan on February 12, 2026, 01:09:58 AM
Well, uhh, wow!

>most definitely Yosemite. So much so that I feel like I should be able to figure out where that climb is.

I was going to ask if you could ID that bridge.
Title: Re: If you can spot AI content today....
Post by: rcjordan on February 12, 2026, 01:17:46 AM
<+>

>human models

See #31
https://www.visualcapitalist.com/ranked-the-jobs-most-exposed-to-generative-ai-according-to-microsoft/
Title: Re: If you can spot AI content today....
Post by: Rupert on February 12, 2026, 06:37:36 AM
Quotecustomer-facing roles such as sales representatives, customer service agents, telemarketers, and concierges

Interesting combination, very tech-based view. Putting "Sales" with "concierges" feels off to me.
Title: Re: If you can spot AI content today....
Post by: ergophobe on February 12, 2026, 09:26:54 PM
>  ID that bridge

So most features are immediately recognizable. It's basically a photo of the valley with a climbing put into the foreground - Half Dome (obviously), but also Royal Arches, Washington Column, North Dome, Cloud's Rest, Leidig Meadow, Ahwahnee Meadow, the lower toe of Lower Brother, and, to answer your question, Swinging Bridge (a footbridge).

On the Valley floor, it's not quite as flawless. Leidig Meadow does not have a prominent dirt road like that. I also *think* though I'm not sure, that the beach in the lower right is as large as shown, though again, I would have to look.

I do not think this is a *possible* view. If you're on Manure Pile Buttress, you're too low and and far to the right.
https://www.mountainproject.com/photo/114349553/yeeeeahhhh-topping-out-nutcracker

If you are on the East Buttress of El Cap, there's too much in your way to get that nice view of the meadows and the toe of Lower Brother. It might be possible.

In this image, I'm leading out to the right around that ridge. It's possible that when I get out there, I'm at the right elevation and would see all the things in that photo. I'll have to look next time I'm up there.

But it is *really* believable that that photo is taken from somewhere near the East Buttress of El Cap, though the route that I am on is the furthest route of any route on El Cap. So the climb is invented, but really artfully so. You have to know Yosemite quite well to say there is no route with that view and I'm still not 100% confident that I can say that.


>  Putting "Sales" with "concierges" feels off to me.

Not to me. The really good sales people and concierges are extremely hard to replace. The crappy ones are easy to replace. Most people in sales and concierge services are not experts in any way.
Title: Re: If you can spot AI content today....
Post by: ergophobe on February 12, 2026, 09:34:50 PM
Here's someone just a bit further around the corner than the shot of me
https://www.mountainproject.com/photo/123086869/pitch-9-knobs

As you can see, you would still need to get quite a bit further around the corner and then have a photographer who had pre-rigged to be in the right spot with a decent telephoto to compress the depth of field. I still don't think you get that view.

Here's someone who has topped out El Cap and is coming down the East Ledges descent. You can see that by that point you're now too far left and too high for the view in the photo
https://www.lamountaineers.org/NAC/browserf/climbs/elcapnos/hi/11U05.jpg

Conclusion: that photo is impossible, but I have to research and look through photos and so forth to be sure.

I'm speaking of the landscape view. The portrait view has some subtle problems with the view (generally just way too compressed), and of course the obvious problems with the climber
Title: Re: If you can spot AI content today....
Post by: rcjordan on February 12, 2026, 10:12:32 PM
> *really* believable that that photo is taken from somewhere near the East Buttress of El Cap

A photog with a drone assistant to 'lean out' perhaps?  Anyway, good enough for a non-climber mag.

>really good sales people and concierges are extremely hard to replace. The crappy ones are easy to replace

Same for receptionists ...which is another name for concierge, I guess

Title: Re: If you can spot AI content today....
Post by: ergophobe on February 12, 2026, 10:48:56 PM
> receptionists

My doctoral advisor was helping me through a problem at the university of some sort. I forget the problem, but I wasn't getting anywhere until he sent me to the *right* admin. He quipped, "The university can survive just fine getting rid of a hundred professors or a dozen deans, but there is a cadre of middle-aged women* who actually know how the university works and if a couple dozen of them left at once, the university would collapse in days."

* and, yes, in the 1980s, it was overwhelmingly women in these roles and the middle-aged ones were the ones with the experience and knowledge that could not be replaced.


> drone assistant

It's possible. But I would have to actually do the experiment (which would be against the law BTW) to know. And that's the staggering thing. All the time I've spent here and there in Yosemite, including 4-5 times up the East Buttress and a dozen or so times down the East Ledges and, without going there with a drone and testing, I can't say for sure whether or not that pic is possible to take IRL.

Astounding progress since August
Title: Re: If you can spot AI content today....
Post by: rcjordan on February 13, 2026, 12:32:56 AM
>aged women* who actually know how the university X works

I'll insert here that there's a story here in th3core about the elder woman who ran the Selective Service office using her know-how to keep me (draft lottery #26) from being VietNam war fodder in 1972.
Title: Re: If you can spot AI content today....
Post by: Rupert on February 13, 2026, 09:54:48 AM
I had to look up the term 'concierge,' as my understanding of the role was clearly off. Fun fact from Gemini

QuoteThe "Les Clefs d'Or"

You might see a concierge wearing crossed golden keys on their lapels. This signifies they belong to Les Clefs d'Or (The Golden Keys), an elite international network of the world's best concierges. If they have those keys, they have a massive network of contacts to make the "impossible" happen