Group items matching

in title, tags, annotations or url

Mensch said his new model cost less than €20 million, the equivalent of roughly $22 million, to train. By contrast OpenAI Chief Executive Sam Altman said last year after the release of GPT-4 that training his company’s biggest models cost “much more than” $50 million to $100 million.

Brave Software made a free, open-source model from Mistral the default to power its web-browser chatbot, said Brian Bondy, Brave’s co-founder and chief technology officer. He said that the company finds the quality comparable with proprietary models, and Mistral’s open-source approach also lets Brave control the model locally.

“We want to be the most capital-efficient company in the world of AI,” Mensch said. “That’s the reason we exist.” 

Mensch joined the Google AI unit then called DeepMind in late 2020, where he worked on the team building so-called large language models, the type of AI system that would later power ChatGPT. By 2022, he was one of the lead authors of a paper about a new AI model called Chinchilla, which changed the field’s understanding of the relationship among the size of an AI model, how much data is used to build it and how well it performs, known as AI scaling laws.

Mensch took a role lobbying French policymakers, including French President Emmanuel Macron, against certain elements of the European Union’s new AI Act, which Mensch warned could slow down companies and would, in his view, do nothing to make AI safer. After changes to the text in Brussels, it will be a manageable burden for Mistral, Mensch says, even if he thinks the law should have remained focused on how AI is used rather than also regulating the underlying technology.  

For Mensch and his co-founders, releasing their initial AI systems as open source that anyone could use or adapt free of charge was an important principle. It was also a way to get noticed by developers and potential clients eager for more control over the AI they use

Mistral’s most advanced models, including the one unveiled Monday, aren’t available open source. 

he trickiest question may be how to prevent abuse. AI generators have technological boundaries, but not morals, and it’s relatively easy for users to trick them into creating content that depicts violence, rape, sex with children or a celebrity — or even a crush from work who never consented to appear

In some cases, the engines themselves are trained on porn images whose subjects didn’t explicitly agree to the new use. Currently, no federal laws protect the victims of nonconsensual deepfakes.

Adult entertainment is a giant industry accounting for a substantial chunk of all internet traffic: Major porn sites get more monthly visitors and page views than Amazon, Netflix, TikTok or Zoom

The industry is a habitual early adopter of new technology, from VHS to DVD to dot com. In the mid-2000s, porn companies set up massive sites where users upload and watch free videos, and ad sales foot the bills.

At last year’s AVN conference, Steven Jones said his peers looked at him “like he was crazy” when he talked about AI opportunities: “Nobody was interested.” This year, Jones said, he’s been “the belle of the ball.”

He called up his old business partner, and the two immediately spent about $550,000 securing the web domains for porn dot ai, deepfake dot com and deepfakes dot com, Jones said. “Lightspeed” was back.

One major model, Stable Diffusion, shares its code publicly, and some technologists have figured out how to edit the code to allow for sexual images

What keeps Jones up at night is people trying to use his company’s tools to generate images of abuse, he said. The models have some technological guardrails that make it difficult for users to render children, celebrities or acts of violence. But people are constantly looking for workarounds.

So with help from an angel investor he will not name, Jones hired five employees and a handful of offshore contractors and started building an image engine trained on bundles of freely available pornographic images, as well as thousands of nude photos from Jones’s own collection

Users create what Jones calls a “dream girl,” prompting the AI with descriptions of the character’s appearance, pose and setting. The nudes don’t portray real people, he said. Rather, the goal is to re-create a fantasy from the user’s imagination.

The AI-generated images got better, their computerized sheen growing steadily less noticeable. Jones grew his user base to 500,000 people, many of whom pay to generate more images than the five per day allotted to free accounts, he said. The site’s “power users” generate AI porn for 10 hours a day, he said.

Jones described the site as an “artists’ community” where people can explore their sexualities and fantasies in a safe space. Unlike some corners of the traditional adult industry, no performers are being pressured, underpaid or placed in harm’s way

And critically, consumers don’t have to wait for their favorite OnlyFans performer to come online or trawl through Pornhub to find the content they like.

Next comes AI-generated video — “porn’s holy grail,” Jones said. Eventually, he sees the technology becoming interactive, with users giving instructions to lifelike automated “performers.” Within two years, he said, there will be “fully AI cam girls,” a reference to creators who make solo sex content.

It costs $12 per day to rent a server from Amazon Web Services, he said, and generating a single picture requires users to have access to a corresponding server. His users have so far generated more than 1.6 million images.

Copyright holders including newspapers, photographers and artists have filed a slew of lawsuits against AI companies, claiming the companies trained their models on copyrighted content. If plaintiffs win, it could cut off the free-for-all that benefits entrepreneurs such as Jones.

But Jones’s plan to create consumer-friendly AI porn engines faced significant obstacles. The companies behind major image-generation models used technical boundaries to block “not safe for work” content and, without racy images to learn from, the models weren’t good at re-creating nude bodies or scenes.

Jones said his team takes down images that other users flag as abusive. Their list of blocked prompts currently contains 1,000 terms including “high school.”

“I see certain things people type in, and I just hope to God they’re trying to test the model, like we are. I hope they don’t actually want to see the things they’re typing in.

Peter Acworth, the owner of kink dot com, is trying to teach an AI porn generator to understand even subtler concepts, such as the difference between torture and consensual sexual bondage. For decades Acworth has pushed for spaces — in the real world and online — for consenting adults to explore nonconventional sexual interests. In 2006, he bought the San Francisco Armory, a castle-like building in the city’s Mission neighborhood, and turned it into a studio where his company filmed fetish porn until shuttering in 2017.

Now, Acworth is working with engineers to train an image-generation model on pictures of BDSM, an acronym for bondage and discipline, dominance and submission, sadism and masochism.

Others alluded to a porn apocalypse, with AI wiping out existing models of adult entertainment.“Look around,” said Christian Burke, head of engineering at the adult-industry payment app Melon, gesturing at performers huddled, laughing and hugging across the show floor. “This could look entirely different in a few years.”

But the age of AI brings few guarantees for the people, largely women, who appear in porn. Many have signed broad contracts granting companies the rights to reproduce their likeness in any medium for the rest of time

Not only could performers lose income, Walters said, they could find themselves in offensive or abusive scenes they never consented to.

Lana Smalls, a 23-year-old performer whose videos have been viewed 20 million times on Pornhub, said she’s had colleagues show up to shoots with major studios only to be surprised by sweeping AI clauses in their contracts.

“This industry is too fragmented for collective bargaining,” Spiegler said. “Plus, this industry doesn’t like rules.”

Rising inflation has led to surging prices for groceries, cars and many essentials. The cost of a college education has skyrocketed in two decades, with the average student debt for federal loans at $37,000, according to the Education Department. Overall debts for Americans in their 30s jumped 27% from late 2019 to early 2023—steeper than for any other age group.

home prices have risen considerably, while wages haven’t kept pace.

“What Dave Ramsey really misses is any kind of social context,” says Morgan Sanner, a

She began paying off $48,000 in student loans (a Ramsey do) and also took out a loan to buy a 2016 Honda (a Ramsey don’t). Her rationale was that it was safer to pay extra for a more reliable car than a junker she could buy with cash. S

he feels these sorts of real-life decisions don’t factor into his advice.

When she saw a comment from Ramsey online about how people receiving pandemic stimulus payments were “pretty much screwed already,” Israel felt it came across as shaming people. The pandemic shutdowns ended a decadelong economic expansion for Black Americans, a disproportionate number of whom lost their jobs and relied on those checks.

“Moralizing financial decisions is very damaging to marginalized groups,” says Israel, who is Black.

Many young adults scratch their heads over his advice that people should let their credit scores dwindle and die.

People need a good credit score, says Mandy Phillips, a 39-year-old residential mortgage loan originator in Redding, Calif. She uses TikTok and other social media to educate millennials and Gen Z about home buying. Scores are vital when applying for mortgages and rentals.

She also takes issue with Ramsey’s advice to only obtain a home loan if you can take out a 15-year fixed-rate mortgage with a down payment of at least 10%. Few younger buyers can pay the large monthly bills of shorter-term mortgages.

“That may have worked years ago in the ’80s and ’90s, but that’s not something that is achievable for the average American,” Phillips says.

Housing is a particularly hot-button topic. He advises people to only buy a house with their lawfully wedded spouse. Yet many young adults are pooling their finances with partners, friends or roommates to buy their first homes. 

Ramsey is perhaps best known for advocating a “debt snowball method”: People with multiple loans pay off the smallest balances first, regardless of interest rate. As you knock out each loan, he says, the money you have to put toward larger debt snowballs. Seeing small wins motivates people to keep going, he says.Conventional economic theory would be to pay off the highest-interest loans first, says James Choi, a finance professor at the Yale School of Management, who has studied the advice of popular finance gurus.

Ramsey’s save-not-spend message sounds logical, young adults say. It’s his all-or-nothing approach that doesn’t work for them.

Kate Hindman, a 31-year-old administrative assistant in Pasadena, Calif., who has taken an anti-Ramsey stance on TikTok, ended up with $30,000 in credit-card debt after she and her husband faced income-reducing job changes. They’ve since turned it into a consolidation loan with an 8% interest rate and pay about $1,200 a month.

She wonders if the debt aversion is generational. Perhaps younger people are less willing to make huge sacrifices to be debt-free. Maybe carrying some amount of debt forever is a new normal.