The AI dooooooom thread

[Barrister]

Why is it a money loser?

I assume insufficient sales to cover publishing.

I believe the books in question are either e-books, or print-on-demand books.

https://en.wikipedia.org/wiki/Print_on_demand

So even if you sell a handful of books you're still profiting on each one.

[Syt]

Why is it a money loser?

I assume insufficient sales to cover publishing.

Publishing on Amazon or Smashwords doesn't cost you anything, and if you generate all content yourself, it only costs you time - and with AI generating the content for you, you just need to format it in the ebook format of choice and upload it.

[Sheilbh]

They're already doing it, so it must make sense.

Presumably using AI prompts to generate 500 "books" on different topics, uploading them, and using botnets to push them to the top of search rankings is something that is relatively easy to automate.

At which point the money from suckers buying any of those "books" (that it costs pennies to generate) is pure profit.

Yeah - and it's similar with online content.

You have an increasing long tail of the internet which is made for advertising sites (often AI generated, and nonsense) but with sellable ad inventory and that attracts clicks. Ad sales are often priced best for the ad slots that have the most personalised/developed profile of the individual at the end and allow the most functionalities (both of which require intrusive tracking). That means that at the minute the industry is spending money on sites with basically no compliance with any privacy laws, that are just created for the purpose of attracting eyes on adverts and just a churn with no permanence.

Admittedly this is possibly the logical end point of online advertising given that it is an industry which is largely based on fraud which is now being accelerated by AI. But it is not good for individuals, not good for publishers of real content and not good for advertisers. And ultimately as long as the things that agencies really prioritise (because it's what advertisers insist on) are targeting and functionalities, it's very difficult to move it in a different direction.

And obvioiusly those sites are being scraped and included in future models - particularly ones reliant on webcrawlers that quality media companies are increasingly blocking because they want to get paid for their content.

[celedhring]

Just noticed chatGPT can talk too. Predictably, it sounds like a smug asshole.

EDIT: Ah, there's different voices. In Catalan "Juniper" has the most incredibly Barcelona posh accent I can think of. Shit's scary.

[grumbler]

Why is it a money loser?

I assume insufficient sales to cover publishing.

The costs of self-publishing are primarily the costs of formatting the book properly and perhaps a bit of advertising.  The formatting issue isn't an issue with AI.

[garbon]

After I performed my first task for work with ChatGPT (at a spead I'd never have managed on my own), I then asked it to create me a story about a woman who won the marathon against underwhelming odds in the style of Jane Austen.

In the quaint village of Longbourne, nestled amidst the rolling English countryside, there lived a spirited young woman named Elizabeth Bennet. Elizabeth was known for her intelligence, wit, and independent spirit, qualities not always appreciated by the more conventional society around her.

One fine autumn day, Longbourne was abuzz with excitement as news spread that a marathon race was to be held in nearby Meryton. This event, unusual for its time, attracted athletes from far and wide, including the dashing Mr. Darcy, whose impressive athleticism and serious demeanor had captured the attention of many a young lady.

Intrigued by the prospect of such a spectacle, Elizabeth found herself drawn to the event despite the protests of her mother, Mrs. Bennet, who could not understand her daughter's fascination with "such unseemly pursuits." Undeterred, Elizabeth resolved to attend, her curiosity piqued by the challenge and the chance to witness Mr. Darcy's prowess firsthand.

On the morning of the marathon, a colorful crowd gathered at the starting line in Meryton. Elizabeth, dressed in a practical but elegant running gown, stood amongst the spectators, her eyes fixed on the determined faces of the runners. Mr. Darcy, tall and composed, exuded an air of confidence that both intrigued and annoyed Elizabeth.

As the race commenced, Elizabeth's attention quickly shifted from Mr. Darcy to a lesser-known competitor, Miss Jane Bingley, a gentlewoman of refined manners and surprising athletic ability. Miss Bingley's grace and stamina caught Elizabeth's eye, and she found herself silently cheering for this unexpected heroine of the track.

The course wound through the picturesque countryside, and as the hours passed, the runners faced various challenges—hilly terrain, unpredictable weather, and their own physical limitations. Elizabeth observed with admiration as Miss Bingley, with quiet determination, steadily gained ground on the leading pack, which included Mr. Darcy.

As the marathon neared its conclusion, Elizabeth's excitement grew. Miss Bingley, previously an underdog in the eyes of many, was now closing in on victory. With a burst of speed that surprised even herself, Miss Bingley crossed the finish line, cheered on by a jubilant crowd.

In that moment, Elizabeth realized the significance of what she had witnessed. It was not merely a race won by physical prowess but a triumph of spirit and determination. Miss Bingley's victory challenged societal norms and proved that women were capable of achieving greatness beyond the confines of traditional expectations.

As Elizabeth made her way back to Longbourne that evening, her heart was filled with a newfound sense of possibility. She had witnessed a woman overcoming underwhelming odds with grace and fortitude—a testament to the indomitable spirit that resided within each of them.

Inspired by Miss Bingley's example, Elizabeth vowed to pursue her own dreams with unwavering determination, undeterred by the obstacles that society might place in her path. For in the fields of Longbourne, amidst the whispers of the autumn breeze, a new kind of heroine had emerged—one whose story would be told and celebrated for generations to come.

[Sheilbh]

Not the normal AI Doom piece - but this story is incredible. I'd heard reporters worry about audio fakes but honestly I hadn't really thought about the use of AI for this sort of day-to-day, real world maliciousness (also I have some questions for the teacher who immediately shared it with a student to spread ):

Ex-athletic director accused of framing principal with AI arrested at airport with gun
Kristen Griffith and Justin Fenton
4/25/2024 8:44 a.m. EDT, Updated 4/25/2024 5:58 p.m. EDT
The principal of Pikesville High School was investigated after audio purporting to be his voice circulated on social media. Police have charged the former athletic director who they say faked the recording using artificial intelligence software.

Baltimore County Police arrested Pikesville High School's former athletic director Thursday morning and charged him with using artificial intelligence to impersonate Principal Eric Eiswert, leading the public to believe Eiswert made racist and antisemitic comments behind closed doors.

Dazhon Darien, 31, was apprehended as he attempted to board a flight to Houston at BWI Airport, Baltimore County Police Chief Robert McCullough said at a news conference Thursday afternoon. Darien was stopped for having a gun on him and airport officials saw there was a warrant for his arrest. Police said they did not know whether Darien was trying to flee.

Darien was charged with disrupting school activities after investigators determined he faked Eiswert's voice and circulated the audio on social media in January, according to the Baltimore County State's Attorney's Office. Darien's nickname, DJ, was among the names mentioned in the audio clips authorities say he faked.

"The audio clip ... had profound repercussions," police wrote in charging documents. "It not only led to Eiswert's temporary removal from the school but also triggered a wave of hate-filled messages on social media and numerous calls to the school. The recording also caused significant disruptions for the PHS staff and students."

Police say Darien made the recording in retaliation after Eiswert initiated an investigation into improper payments he made to a school athletics coach who was also his roommate. Darien is also charged with theft and retaliating against a witness.

Darien was allowed release on $5,000 bond and waived an attorney at an initial court appearance, according to court records. Attempts to reach him by phone and at his home were unsuccessful.

Eiswert's voice, which police and AI experts believe was simulated, made disparaging comments about Black students and the surrounding Jewish community and was widely circulated on social media.

Questions about the audio's authenticity quickly followed. Police wrote in charging documents that Darien had accessed the school's network on multiple occasions in December and January searching for OpenAI tools, and used "Large Language Models" that practice "deep learning, which involves pulling in vast amounts of data from various sources on the internet, can recognize text inputted by the user, and produce conversational results." They also connected Darien to an email account that had distributed the recording.

Many current and former students believed Eiswert was responsible for the offensive remarks, while former colleagues denounced the audio and defended Eiswert's character. Eiswert himself has denied making those comments and said the comments do not align with his views.

The audio, posted to the popular Instagram account murder_ink_bmore, prompted a Baltimore County Public Schools and Baltimore County Police investigation. Eiswert has not been working in the school since the investigation began.

The voice refers to "ungrateful Black kids who can't test their way out of a paper bag" and questions how hard it is to get those students to meet grade-level expectations. The speaker uses names of people who appear to be staff members and says they should not have been hired, and that he should get rid of another person "one way or another."

"And if I have to get one more complaint from one more Jew in this community, I'm going to join the other side," the voice said.

Darien was being investigated as of December in a theft investigation that had been initiated by Eiswert. Police say Darien had authorized a $1,916 payment to the school's junior varsity basketball coach, who was also his roommate, under the pretense that he was an assistant girls soccer coach. He was not, school officials said.

Eiswert determined that Darien had submitted the payment to the school payroll system, bypassing proper procedures. Darien had been notified of the investigation, police said.

Police say the clip was received by three teachers the night before it went viral. The first was Darien; a third said she received the email and then got a call from Darien and teacher Shaena Ravenell telling her to check her email. Ravenell told police that she had forwarded the email to a student's cell phone, "who she knew would rapidly spread the message around various social media outlets and throughout the school," and also sent it to the media and the NAACP, police said.

She did not mention receiving it from Darien until confronted about his involvement. Ravenell has not been charged with a crime and could not immediately be reached for comment.

Both Darien and Ravenell have submitted their resignations to the school system, according to an April 16 school board document. The resignations are dated June 30.

Baltimore County Public Schools Superintendent Myriam Rogers said school system officials are recommending Darien's termination. She would not say, however, if the other employees named in the charging documents, including Ravenell, are still working at the school.

Rogers in January called the comments "disturbing" and "highly offensive and inappropriate statements about African American students, Pikesville High School staff, and Pikesville's Jewish community."

Rogers said Kyria Joseph, executive director for secondary schools, and George Roberts, a leadership consultant for the school system, have been running Pikesville High School since the investigation started. They will continue to do so for the remainder of the year. She said they will work with Eiswert to determine his duties for next school year.

Billy Burke, head of the Council of Administrative & Supervisory Employee, the union that represents Eiswert, was the only official to suggest the audio was AI-generated.

Burke said he was disappointed in the public's assumption of Eiswert's guilt. At a January school board meeting, he said the principal needed police presence at his home because he and his family had been harassed and threatened. Burke had also received harassing emails, he said at the time.

"I continue to be concerned about the damage these actions have caused for Principal Eiswert, his family, the students and staff of Pikesville High School, and the Black and Jewish community members," Burke said in a statement on Thursday. "I hope there is deliberate action to heal the trauma caused by the fake audio and that all people can feel restored."

Police said the school's front desk staff was "inundated with phone calls from parents and students expressing concern and disparaging remarks toward school staff and administrators." The flood of calls made it difficult to field phone calls from parents trying to make arrangements for their children and other school functions, officials told police.

"The school leadership expressed that staff did not feel safe, which required an increase in police presence at the school to address safety concerns and fears," police said.

Teachers, under the impression the recording was authentic, "expressed fears that recording devices could have been planted in various places in the school," police said.

"The recording's release deeply affected the trust between teachers and the administration," police said. "One individual shared that they fielded sensitive phone calls in their vehicle in the parking lot instead of speaking in school."

"Hate has no place and no home in Baltimore County," said Johnny Olszewski Jr., the Baltimore County executive.

He called the developments of AI "deeply concerning" and that it's important for everyone to remain vigilant for anyone using the technology for malicious reasons. There should also be more investment in technology that identifies any inauthentic recording made with AI, he said.

Experts in detecting audio and video fakes told The Banner in March that there was overwhelming evidence the voice is AI-generated. They noted its flat tone, unusually clean background sounds and lack of consistent breathing sounds or pauses as hallmarks of AI. They also ran the audio through several different AI-detection techniques, which consistently concluded it was a fake, though they could not be 100% sure.

The police also sought the expertise of two professors familiar with AI detection to assist in their investigation. Catalin Grigoras, a forensic analyst and professor at the University of Colorado Denver, concluded that the "recording contained traces of AI-generated content with human editing after the fact, which added background noises for realism," the charging documents stated.

Hany Farid from the University of California, Berkeley, who's also an expert in forensic analysis, determined "the recording was manipulated, and multiple recordings were spliced together," according to the documents.

AI voice-generation tools are now widely available online, and a single minute's recording of someone's voice can be enough to simulate it with a $5-a-month AI tool, the Nieman Journalism Lab reported in February.

There are few regulations to prevent AI imitations, called deepfakes, and few perpetrators are prosecuted.

Cindy Sexton, president of the Teachers Association of Baltimore County, said AI should be a concern for everyone, especially educators.

She said the National Education Association is working to address their concerns, but in the meantime, she's not sure what else should be done.

"We have to do something as a society, but 'what is that something' is of course the big question," Sexton said

Baltimore County State's Attorney Scott Shellenberger said this is the first time this type of case has been taken up by the district. And it's one of the first his office was able to find around the nation.

There were some legal statutes they used that were "right on point," he said, but the charge of disrupting school activities only carries a six-month sentence.

"It seems very clear to me that we may need to make our way down to Annapolis in the legislature next year to make some adaptions to bring the law up to date with the technology that was being used," he said.

Baltimore Banner staff writers Cody Boteler and Kaitlin Newman contributed to this report.

Correction: This story has been updated to correct the spelling of Hany Farid's name.

Edit: The clip. Obviously I don't know the guy and now I listen for it the lack of background noise is noticeable - but if I heard this, I would have no idea:
https://x.com/Phil_Lewis_/status/1747708846942851493

[Josquius]

Some people have been experimenting with using official corporate approved  AI at work with tasks such as with writing up notes from research.
The results are....less  than impressive. The AI always misses critical nuances.
Like there was one example where somebody asked about the timelines for something, they were told it is planned but we can't give you a timeline yet, and replied "OK right."
The AI's summary?
They're enthusiastic about the lack of timelines.
...
err....
And this isn't an exceptional case, lots of things like this that can't just be put down to simple explanations like not hearing the sarcasm in a tone.

I'm wondering whether a problem might lie in the AI being trained heavily on American data thus it just completely screws up when faced with other English speakers?- we're speaking the same language so it thinks it can understand...but not really.
Could be quite the great topic for someone in academia I think.

I have noticed however that it does seem to do a better job of cobbling together readable transcripts than most pre-existing tools for this. Though again, the false confidence it displays means where there are errors they are much greater than they would be otherwise.

[garbon]

It is known that sarcasm has not yet reached America.

[crazy canuck]

Some people have been experimenting with using official corporate approved  AI at work with tasks such as with writing up notes from research.
The results are....less  than impressive. The AI always misses critical nuances.
Like there was one example where somebody asked about the timelines for something, they were told it is planned but we can't give you a timeline yet, and replied "OK right."
The AI's summary?
They're enthusiastic about the lack of timelines.
...
err....
And this isn't an exceptional case, lots of things like this that can't just be put down to simple explanations like not hearing the sarcasm in a tone.

I'm wondering whether a problem might lie in the AI being trained heavily on American data thus it just completely screws up when faced with other English speakers?- we're speaking the same language so it thinks it can understand...but not really.
Could be quite the great topic for someone in academia I think.

I have noticed however that it does seem to do a better job of cobbling together readable transcripts than most pre-existing tools for this. Though again, the false confidence it displays means where there are errors they are much greater than they would be otherwise.

It does not think, and it does not understand.  That is the much more simple explanation for the errors you are seeing.

[Tamas]

It does not think, and it does not understand. 

This should be an important message about these "AIs" but there are too many people benefiting from the doom and gloom hype machine.

[grumbler]

It is known that sarcasm has not yet reached America.

Indeed, the language does not even have a word for sarcasm.

[Barrister]

So there's been an interesting story out there recently: Chat GPT releases a new voice mode that strongly resembles actress Scarlet Johansson's voice.  Johansson voiced an AI in the movie Her a number of years ago.  Johansson was asked to provide the voice for Chat GPT but refused.

Is this a violation of Johansson's rights though?  It does feel creepy - but it's not like voice impersonator's haven't been a thing for years.  I mean Rich Little made a career of it for decades.

And it extends more broadly to AI in general.  These LLMs are all trained on existing material.  Picture AIs are perhaps most interesting (at least to me).  As a person you are of course free to go to all the art galleries of the world and study all of the existing pictures to get "inspired" by them, and no one would call "foul" over doing so as long as you don't directly copy.

But it does seem unseemly - either voice, text, or art - when LLM-trained AIs then interpret existing material to create derivative works.

I'm not sure where I come down on this.

[The Minsky Moment]

Derivative works are part of the copyright bundle of rights under the US law. My understanding is that the AI models are banking on the contention that their activity is sufficiently transformative to qualify as fair use. But the last Supreme Court ruling on that doctrine limited it.

[Barrister]

Derivative works are part of the copyright bundle of rights under the US law. My understanding is that the AI models are banking on the contention that their activity is sufficiently transformative to qualify as fair use. But the last Supreme Court ruling on that doctrine limited it.

But coming back to a non-AI example - the "Blurred Lines" decision, where Robin Thicke and Pharrell Williams were sued by the estate of MArvin Gaye.  The song in question didn't copy any direct chords, melodies or words - but the jury found it infringed Gaye's copyright (and it was upheld on appeal).

That seemed then, and seems now, like a stupid decision.  You can't copyright the "feel" of a song.

But how different is that from AIs?