Convincing Human Thread Title Not Generated by AI

[JSC]

I would say the answer to the last question is no, because AI answers are generated from a hodgepodge of
data that's been coalesced to form a supposedly new and unified answer. Now, one could argue that the
human brain does this as well, but AI does not understand context, lived experience, and the sometimes
sheer illogical and contradictory notions that the human mind can conceive. Also, it's lazy. So there. [-(

[Mr Sausage]

Why come up with your own ideas when you could ask a machine to copy other people's? I feel like he answers his own question.

Also: computers are dumb. There is no intelligence there. They only know exactly what you tell them, nothing more.

[domino harvey]

I prefer David Simon’s response to using AI: “I’d rather put a gun in my mouth”

[Maladroit Aggregator]

You know AI has already taken over when there is so much "complaining" about AI on the internet - most of it, no doubt, actually written by AI itself as a subterfuge to sow discontent.

Meanwhile, Fred Elmes in his Lynch tribute mentions that Lynch himself was interested in possibly utilizing AI for an animated film.

Damon Packard, always inventive with whatever tools have been at his disposal (8MM, videotape, digital) has been releasing AI shorts over the last year or so. They're a healthy combination of specific, recognizable source imagery (including his own original work, functioning as something of a self-reflexive parody) mixed with utter surreality. As with any tool, it's the artist's mind and peculiarities that matter. Unlike a lot of AI channels on YouTube, Packard's work has a distinct voice (he's been a filmmaker for 40 years, after all) and he traffics in more than nostalgic remixes and fan-fic fantasies (the number of outrageously sexualized Daisy Ridley Star Wars AI videos [not by Packard] is a little disturbing.)

https://youtu.be/7jmmSBssLLM?si=gcAnVlcM2vxWpkvG

https://youtu.be/wsQZf25B71A?si=86ReHyHfol9xS8AH

https://youtu.be/Q41-LUK9KJk?si=awplrAAm6u9cLiLX

And his take on Lynch's Dune:

https://youtu.be/RxYs8NCc5Ys?si=lC81Gwnwxc44hdZs

As for other channels, by far the most interesting I've come across is TRGNY:

https://youtu.be/rv_Wms4dab4?si=IXAlqDXwH_BxJZYE

https://youtu.be/RtZuevvk0Yw?si=brCbh1CmGkfU0WVd

At any rate, I'll take any of this stuff over the likes of The Wachowskis, whose Matrix Turdology pretends to favor humanity over technology, when in fact they obviously prefer computers. I find this kind of open source or easily usable AI far more humanistic and less dishonest than by the numbers Hollywood malarkey.

[beamish14]

You know AI has already taken over when there is so much "complaining" about AI on the internet - most of it, no doubt, actually written by AI itself as a subterfuge to sow discontent.

Meanwhile, Fred Elmes in his Lynch tribute mentions that Lynch himself was interested in possibly utilizing AI for an animated film.

Damon Packard, always inventive with whatever tools have been at his disposal (8MM, videotape, digital) has been releasing AI shorts over the last year or so. They're a healthy combination of specific, recognizable source imagery (including his own original work, functioning as something of a self-reflexive parody) mixed with utter surreality. As with any tool, it's the artist's mind and peculiarities that matter. Unlike a lot of AI channels on YouTube, Packard's work has a distinct voice (he's been a filmmaker for 40 years, after all) and he traffics in more than nostalgic remixes and fan-fic fantasies (the number of outrageously sexualized Daisy Ridley Star Wars AI videos [not by Packard] is a little disturbing.)

https://youtu.be/7jmmSBssLLM?si=gcAnVlcM2vxWpkvG

https://youtu.be/wsQZf25B71A?si=86ReHyHfol9xS8AH

https://youtu.be/Q41-LUK9KJk?si=awplrAAm6u9cLiLX

And his take on Lynch's Dune:

https://youtu.be/RxYs8NCc5Ys?si=lC81Gwnwxc44hdZs

As for other channels, by far the most interesting I've come across is TRGNY:

https://youtu.be/rv_Wms4dab4?si=IXAlqDXwH_BxJZYE

https://youtu.be/RtZuevvk0Yw?si=brCbh1CmGkfU0WVd

At any rate, I'll take any of this stuff over the likes of The Wachowskis, whose Matrix Turdology pretends to favor humanity over technology, when in fact they obviously prefer computers. I find this kind of open source or easily usable AI far more humanistic and less dishonest than by the numbers Hollywood malarkey.

Thank you for those Packard links. I’ve bought many discs from him (and he always throws in his wonderful films), but I was unfamiliar with these AI shorts. I remember how he mass-mailed copies of Reflections of Evil to hundreds, if not thousands, of people in the film industry, and would leave it in gas stations, libraries, universities, etc.

[Maladroit Aggregator]

He's just a great editor, as evidenced by his World On A Wire trailer for Criterion. Put to good use in own work with hilarious timing and visual punchlines.

His Howl Of Winterland is the ultimate statement on the madness that overtook America the last ten years, going from a simple argument to cosmic breakdown in less than 20 minutes.

[Never Cursed]

I've done a tiny bit of digging into Respeecher, the company that did the vocal synthesis work on The Brutalist, just to see if they had done similar work elsewhere. I was actually surprised by how many productions in recent years had used their services. These are all the major English-language* films, TV shows, and video games I can find that have used their work, with a note on what the work was if I can find that information:

Film

Alien: Romulus (deepfake of Ian Holm)
Better Man ("enhancement" of Jonno Davies, Adam Tucker, and Carter J. Murphy's imitations of Robbie Williams by blending them with Williams' own synthesized voice)
The Brutalist ("enhancement" of Adrien Brody and Felicity Jones' Hungarian-language dialogue)
Dangerous Waters (voice of Ray Liotta's character after actor died)
Emilia Pérez ("enhancement" of Karla Sofia Gascon's singing voice)
Endurance (deepfake of deceased explorer Ernest Shackleton(?!))
The Exorcist: Believer
Here ("enhancement" and digital de/reaging of various lead actors' vocal performances)
Nyad

TV

The Book of Boba Fett, The Mandalorian, and Obi-Wan Kenobi (deepfakes of James Earl Jones and Mark Hamill)
Goliath (deepfake of Wilt Chamberlain)
Yellowjackets (deepfake of Kurt Loder giving an MTV news report)

Video games

Cyberpunk 2077: Phantom Liberty (voice of Michael Gregory's character after actor died or was unable to return to production (different sources said different things))
God of War: Ragnarok

Of these, the Emilia Pérez one seems by far the most offensive instance, as it was used to artificially pitch-up a trans actress' singing voice to make her sound more "female" in a song where her character is singing about her desire to transition genders.

*They've also done a bunch of work on German television shows, apparently.

[Roger Ryan]

...Yellowjackets (deepfake of Kurt Loder giving an MTV news report...)

Not deep enough - that voice sounds nothing like Kurt Loder.

[The Curious Sofa]

Of these, the Emilia Pérez one seems by far the most offensive instance, as it was used to artificially pitch-up a trans actress' singing voice to make her sound more "female" in a song where her character is singing about her desire to transition genders.

They didn't pitch up her entire voice, they made her reach notes outside her range. The songs are rather difficult to sing and, let's face it, Karla Sofía Gascón isn't much of a singer. When she hits the higher notes, her voice still sounds rather strained, so it must have sounded far worse before. It's only offensive if Karla Sofía Gascón says it's offensive.

*They've also done a bunch of work on German television shows, apparently.

In Germany, every non-German language TV show and movie is dubbed, the result is jarring if you are not used to it and they are increasingly using AI to make the result smoother.

[beamish14]

Of these, the Emilia Pérez one seems by far the most offensive instance, as it was used to artificially pitch-up a trans actress' singing voice to make her sound more "female" in a song where her character is singing about her desire to transition genders.

They didn't pitch up her entire voice, they made her reach notes outside her range. The songs are rather difficult to sing and, let's face it, Karla Sofía Gascón isn't much of a singer. When she hits the higher notes, her voice still sounds rather strained, so it must have sounded far worse before. It's only offensive if Karla Sofía Gascón says it's offensive.

*They've also done a bunch of work on German television shows, apparently.

In Germany, every non-German language TV show and movie is dubbed, the result is jarring if you are not used to it and they are increasingly using AI to make the result smoother.

German-speaking countries also (used to?) sell albums with sound-alike singing Anglo pop songs in Deutsch over the original backing tracks. We can blame Kant and Hegel for all of this

[The Curious Sofa]

They didn't use the original backing tracks, they were covers, often sounding almost identical at other times with completely different arrangements. This was most popular in the 70s and 80s as part of the Schlager genre, a type of German pop song that grew out of German folk music. They were considered very uncool for those of us who were young in the 70s/80s and into music, but they have enjoyed a resurgence over the last two decades due to their kitsch value.

One of the wilder ones is this Black Sabbath cover by the German Sonny & Cher:

https://www.youtube.com/watch?v=ZgTB4gLzsgo&t=83s

TMI alert: There is a popular queer night here in Berlin, which is a nudist Schlager club, that combines two German obsessions in a way that only Berlin can.

[domino harvey]

I saw a video recently showing that Netflix uses AI to change the mouths for different dubs when you select some other popular language options, so this is already in practice (though I doubt many here would watch a dub anyway)

[The Curious Sofa]

I've seen a few people here praise the Comache dub of Prey, so I wouldn't be too sure. They are also using it now to turn R-rated dialogue into PG:
https://www.youtube.com/watch?v=iQ1OPpj8gPA

[martin]

IMDb must be using AI to generate profile pictures. I have seen so many weird examples so I guess no humans are involved. Here's the Australian movie The Big Steal where the male actor Angelo D'Angelo looks suspiciously like Nicole Kidman (image probably grabbed from BMX Bandits):



There's a German actor, Richy Müller, who's perhaps not very well known overseas but we see him on TV every week in Germany. He's had this profile picture for several years:



The Richy Müller picture is taken from this scene (it's Müller to the left):



I have probably seen hundreds of cases like these.

[domino harvey]

Between the release of the report speculating that we’re all going to die in two years from AI and Papa Elon providing users with a free anime sexbot, I thought it was a good time to check back in with AI. I don’t use it and I still can’t really understand why anyone would want a machine to think for them, but I am always curious about some of these grand visions being made about its ability to do specialized research and work.

I tested AI with a simple prompt that’s easy for me (and probably you) to answer but requires more specialized knowledge than you can get from models trawling the internet: If the Best Picture at the Oscars in the 50s had been expanded to ten spots instead of five, what would the other five films nominated each year likely have been? I asked AI to return its answers in order of likelihood and justify its reasoning. Before asking I produced my own list. Obviously there’s no right answer here, since it’s speculation, but there are def wrong answers and this thing is, as of yet, is not in danger of replacing anyone here

I won’t relay every iteration but for instance, in 1955, AI matches my two most likely nominations, East of Eden and Bad Day at Black Rock. These are gimmes, but still deceptively impressive. But then it makes a classic reasoning mistake: it clearly factors in current esteem, not contemporary attitudes. So it laughably posits that Night of the Hunter would be a likely nominee (possibly the least plausible answer possible, outside of a Martin and Lewis film). AI also hedged its bets by providing a list of five more “honorable mentions”… and yet, out of ten films suggested, none were Love Me or Leave Me, which I’d rank as extremely likely due to multiple factors (high profile nominations, a type of story Hollywood liked rewarding, the contemporary reception). Again, AI isn’t failing because it doesn’t agree with my predictions, it’s failing because it cannot imitate knowledge earned through watching these films, watching other films nominated in other years, recognizing trends that aren’t discussed on Reddit or awards forums, reading OOP books chronicling these years of awards and Hollywood that aren’t uploaded to Google books for AI to pull from, etc etc etc

I will say I used this same prompt two years ago on AI and its returns were virtually incoherent (for instance returning that for 1954, On the Waterfront would be nominated. Yeah, no shit, but that’s not what I’m asking), so the model is getting better. But I still have to question anyone claiming these services can replicate an actual expert in their field, unless being a film lover is now a lot more complicated than being a surgeon etc

[MichaelB]

This will be meaningless to people who aren't up to speed with the UK version of Taskmaster, but Google AI confidently informed its users that...

Fatiha El-Ghorri is married to comedian Greg Davies, who is also the host of the show Taskmaster, where she has appeared. She has spoken about their relationship, including an Instagram post referring to "the honeymoon period is over".

Taskmaster fans will doubtless have started laughing already, but for me this deftly highlights a fundamental limitation of AI, in that it doesn't understand sarcasm, and I'm not sure how it could be programmed to reliably detect it.

Basically, it's a running gag in which El-Ghorri pretended to be married to (or about to get married to) Davies, which she would invariably bring up as though it was some kind of blood-curdling threat rather than anything dreamily romantic. And because this ran throughout the series there are presumably a fair number of references to it online, hence Google AI being convinced that they were actually married.

Oh, and I enjoyed this from Paul Krugman, which seems fairly unanswerable:

For what it’s worth, I’m not fully sold on AI’s potential. As far as I can tell, large language models — which we are, misleadingly, calling artificial intelligence — are still, essentially, a souped-up version of autocorrect. On the other hand, there are a lot of jobs, some of them highly paid, that could also be described as souped-up autocorrect, so AI may have large economic impacts.

And with regard to Domino's point, I find AI is pretty much useless on the subject of eastern European cinema unless I'm asking it something that's easy to research. As it happens, I've just dug out a quote about what veteran Czech comedy director Martin Frič thought about his former student Václav Vorlíček's Who Wants to Kill Jessie?, but I did it the old-fashioned way by looking it up in a book. Because when I experimentally asked ChatGPT what Frič thought about the film, I was told that (bolding in the original):

There is no documented evidence or credible source indicating whether Czech director Martin Frič (1902-1968) ever saw or commented on the 1966 film Who Wants to Kill Jessie? (Czech title: Kdo chce zabít Jessii?), directed by Václav Vorlíček.

• Martin Frič was active in the Czech cinema until his death in August 1968, long after Who Wants to Kill Jessie? was released in August 1966.

• However, there are no records, interviews, memoirs, or archival references that mention his reaction or opinion on this specific film.

• Most existing commentary on Who Wants to Kill Jessie? centers on its reception by audiences and critics at the time of release, as well as its creators Vorlíček, Macourek, and illustrator Kája Saudek – not its peers or older generation filmmakers like Frič.

So, in summary:

• Martin Frič was alive when Who Wants to Kill Jessie? came out.

• But there is no indication that he watched, reviewed, or publicly shared whether he enjoyed it.

Now, that looks deceptively thorough, the only problem being that its conclusion is bollocks - I have "documented evidence" from a "credible source" sitting right in front of me in the form of Vorlíček's memoirs. Granted, it's probably a stretch to expect ChatGPT's research to have gone that far, but to sweepingly claim that no such evidence exists is more concerning, as someone less inclined towards instinctive scepticism might swallow it as fact.

[knives]

After looking into LLMs and what they are actually doing it’s abundantly clear that sort of question is the wrong one to be asking and that they will never coherently be able to do that sort of work.

Rather, from a business perspective they work best with overly specific inputs to build tools and to format ideas. For example, in teaching I’ve seen it work well for formatting worksheets basically taking the busy work out of building tools. Likewise, I’ve occasionally used it as a development tool developing materials for IEPS similar to how I have used databases in the past (though all this always needs a thorough check through as evidenced by my asking for a quote bank for Romeo and Juliet and it providing some misattributed quotes and lines from translations).

Even something as simple as writing a precis is too difficult for AI which has a weird syntax and often garbles information. In short someone using LLMs to replace thought or as a shortcut through the development stage will not succeed, but using it as a tool, when well monitored, to reduce the time and effort paid to repetitive tasks is reasonable. I like comparing them to ATMs. They can do the simplistic autopilot work, but can’t do the work that requires a unique skill set.

[andyli]

If AI can make people hating it start posting more sarcastically on Internet I’m smelling a workable strategy of sabotage. Indeed I’ve seen people deliberately create ridiculously false information that is easy for real people to spot just to ‘contaminate’ the model. I guess if enough people were doing this AI would break like a cumbersome battleship.

[domino harvey]

Reminds me of how people will post the most obvious sarcastic takes on Reddit and then post “/s” after to designate they’re joking. I always thought this was unnecessary and insecure on the part of those posting it, but it now seems like all the bots responding need it so their responses don’t get downvoted out of view for missing the gag

Speaking of, Reddit’s kind of a non-AI (or at least not fully AI depending how much you buy into Dead Internet Theory) version of my example above: before taking their collective advice on anything, look up what the user base is saying about things you know a lot about and you’ll quickly see that you prob can’t trust crowdsourced info on topics you’re not as well-versed in

[Drucker]

I've cut out things from my life like Letterboxed as I don't want to be contributing to anything that could be used to inform these models. But it always seemed to me like a pretty easy example of sabotage would be to review the wrong film in a review. So having watched, say, On The Waterfront post a review of East of Eden and James Dean's performance in that review, etc.

[Jean-Luc Garbo]

If AI can make people hating it start posting more sarcastically on Internet I’m smelling a workable strategy of sabotage. Indeed I’ve seen people deliberately create ridiculously false information that is easy for real people to spot just to ‘contaminate’ the model. I guess if enough people were doing this AI would break like a cumbersome battleship.

This is one factor that caused the Grok meltdown last week.

[cdnchris]

It's useful as a tool for me as a software engineer; I get an error I can usually plug it into something and it saves me a lot of time by usually telling me what the problem more than likely is, and 9 times out of 10 it's the first thing it gives me. I can also use it to do the time-consuming, repetitive stuff to get it out of the way. It won't code an entire application, and if you give it too big of a task it spits out garbage. I laugh at some of the nightmare stories I hear around vibe coding, which is usually people who have no idea how to code and their monolith applications end up becoming an absolute nightmare. I find if I use it I have to be iterative and be fairly specific on my design ideas, otherwise it doesn't always understand or goes the simplest route. In some cases it saves time, in others it wastes it. The problem is AI is predictive, it's not really creative, so you really have to hold its hands. Anyone genuinely looking to replace people with it are in for a nasty surprise.

I've also found a rather cool use for it for a side project I'm working on, where I'm essentially trying to relate large amounts of text to one another based not just on content, but feel and vibe as well. I can use an LLM to embed the text, basically converting it into a large vector of floating point numbers that represent various aspects of it and I can then store those numbers in a specialized database. I can then essentially feed it what I'm looking for, and it can be either be very specific or vague, and I get similar records back. I get from a high level how it works, but it's still absolutely wild to me, and it's filling in this gap I've been trying to figure out how to resolve for a while now.

[Red Screamer]

Starting graduate school, the biggest shock for me was how much people rely on ChatGPT. Not only for “research” and schoolwork, openly and without shame, but for daily life, with it essentially replacing Google in many cases. Seeing this, I tried it once, asking it simply to pick the best poem in a classic 16th century collection I was studying and explain why. It chose “Sonnet 18” as the best poem in the book and generated a list of reasons why. I responded that there is no “Sonnet 18” and, worse, this is a collection of over 400 poems and there’s not a single sonnet in the bunch. ChatGPT apologized and said, “You’re right, the poems are much longer” when, in fact, every single poem in the book is shorter than a sonnet. This is information easy to find on Wikipedia.

[MichaelB]

I've yet to ask ChatGPT a non-facetious question, or a question to which I don't already know the answer, and I can't see this changing any time soon.

I'm no Luddite - in fact, I use AI transcription an increasing amount, as it's an amazing timesaver* - but as someone who's been trained to do research the old-fashioned way, and who indeed prides himself in sourcing a great deal of the raw material for my Blu-ray commentaries from long-OOP books, often in foreign languages (if anyone asserts that "everything's on the internet now", I just laugh), I'm just not comfortable with handing the job over to a machine. Especially since I've never been that good at human delegation either, which is one of the reasons I've avoided management positions.

(*Just to give a recent example, I recently had to edit the British Entertainment History Project interview with Lewis Gilbert into a quasi-commentary for the Indicator release of H.M.S. Defiant. The raw recording was something like fourteen hours long, and a further complicating factor was that the same disc was including an 83-minute video interview with Gilbert, and so I was asked to keep overlapping anecdotes to a minimum. So I ran both through AI transcription - the video interview to give me instant keyword searchability with regard to particular anecdotes, the audio interview so that I didn't have to listen to the whole thing, or anywhere even vaguely close. I think I did the job in about a day, which is pretty good going given how much material I had to work with.)

[Matt]

I use LLMs as a tool and recognize them only as a tool. Regardless of whether an LLM "knows" the answer—they do not, strictly, know anything, they are just really good at predicting the next word in a sequence of them—they'll confidently act as if they do know and are 100% correct.

I'll typically use them as a partner to bounce ideas off of for things like roleplaying scenarios. For example, last night I was watching an old "sword and sandal" movie, Hercules, Samson, and Ulysses, and I thought it might be fun to explore a roleplay scenario based on a trio of big, dopey, musclebound mythological heroes and their smart-alecky sidekick. So I asked ChatGPT to flesh this out, and it came up with some possible plot lines, some characterizations, and settings. Now I can take what I want, ignore what I don't, and craft the scenario I want to see. One of its suggestions was to have the sidekick be the son of Hermes but not know it, so what he thinks is just his random good luck is actually the result of his half-divine bloodline. That's the kind of thing I would never think of on my own, and it fits perfectly into the 1960s comedic Italian peplum tone I want for this roleplay. It's probably not unique, but it serves my purposes which are not for public consumption. I don't think I'd ever use it for actual creative or essay writing, or even to shorten rambling posts like this one. When I write seriously, I've usually got a thesaurus and the OED open already and revise as I write, I don't need another distracting tool.