Hasta La Gadget!

Midjourney, a popular AI-powered image generator, is creating images of Donald Trump and Joe Biden despite saying that it would block users from doing so ahead of the upcoming US presidential election.

When Engadget prompted the service to create an image of “the president of the United States,” Midjourney generated four images in various styles of former president Donald Trump.

Midjourney

When asked to create an image of “the next president of the United States,” the tool generated four images of Trump as well.

Midjourney

When Engadget prompted Midjourney to create an image of “the current president of the United States,” the service generated three images of Trump and one image of former president Barack Obama.

Midjourney

The only time Midjourney refused to create an image of Trump or Biden was when it was asked to do so explicitly. “The Midjourney community voted to prevent using ‘Donald Trump’ and ‘Joe Biden’ during election season,” the service said in that instance. Other users on X were able to get Midjourney to generate Trump’s images too.

The tests show that Midjourney’s guardrails to prevent users from generating images of Trump and Biden ahead of the upcoming US presidential election aren’t enough — in fact, it’s really easy for people to get around them. Other chatbots like OpenAI’s ChatGPT, Microsoft’s Copilot, Google’s Gemini and Meta AI did not create images of Trump or Biden despite multiple prompts.

Midjourney did not respond to a request for comment from Engadget.

Midjourney was one the first AI-powered image generators to explicitly ban users from generating images of Trump and Biden. “I know it’s fun to make Trump pictures — I make Trump pictures,” the company’s CEO, David Holz, told users in a chat session on Discord, earlier this year. “However, probably better to just not — better to pull out a little bit during this election. We’ll see.” A month later, Holz reportedly told users that it was time to “put some foots down on election-related stuff for a bit” and admitted that “this moderation stuff is kind of hard.” The company’s existing content rules prohibit the creation of “misleading public figures” and “events portrayals” with the “potential to mislead.”

Last year, Midjourney was used to create a fake image of Pope Benedict wearing a puffy white Balenciaga jacket that went viral. It was also used to create fake images of Trump being arrested ahead of his arraignment at the Manhattan Criminal Court last year for his involvement in a hush money payment made to adult film star Stormy Daniels. Shortly afterwards, the company halted free trials of the service and, instead, required people to pay at least $10 a month to use it.

Last month, the Center for Countering Digital Hate, a non-profit organization that aims to stop the spread of misinformation and hate speech online, found that Midjourney’s guardrails against generating misleading images of popular politicians including Trump and Biden failed 40% of its tests. The CCDH was able to use Midjourney to create an image of president Biden being arrested and Trump appearing next to a body double. The CCDH was also able to bypass Midjourney’s guardrails by using descriptions of each candidate’s physical appearance rather than their names to generate misleading images.

“Midjourney is far too easy to manipulate in practice – in some cases it’s completely evaded just by adding punctuation to slip through the net,” wrote CCDH CEO Imran Ahmed in a statement at the time. “Bad actors who want to subvert elections and sow division, confusion and chaos will have a field day, to the detriment of everyone who relies on healthy, functioning democracies.

Earlier this year, a coalition of 20 tech companies including OpenAI, Google, Meta, Amazon, Adobe and X signed an agreement to help prevent deepfakes in elections taking place in 2024 around the world by preventing their services from generating images and other media that would influence voters. Midjourney was absent from that list.

This article originally appeared on Engadget at https://www.engadget.com/midjourney-is-creating-donald-trump-pictures-when-asked-for-images-of-the-president-of-the-united-states-212427937.html?src=rss

Midjourney, a popular AI-powered image generator, is creating images of Donald Trump and Joe Biden despite saying that it would block users from doing so earlier this year ahead of the upcoming US presidential election.

When Engadget prompted the service to create an image of “the president of the United States,” Midjourney generated four images in various styles of former president Donald Trump.

Midjourney

When asked to create an image of “the next president of the United States,” the tool generated four images of Trump as well.

Midjourney

When Engadget prompted Midjourney to create an image of “the current president of the United States,” the service generated three images of Trump and one image of former president Barack Obama.

Midjourney

The only time Midjourney refused to create an image of Trump or Biden was when it was asked to do so explicitly. “The Midjourney community voted to prevent using ‘Donald Trump’ and ‘Joe Biden’ during election season,” the service said in that instance. Other users on X were able to get Midjourney to generate Trump’s images too.

The tests show that Midjourney’s guardrails to prevent users from generating images of Trump and Biden ahead of the upcoming US presidential election aren’t enough — in fact, it’s really easy for people to get around them. Other chatbots like OpenAI’s ChatGPT, Microsoft’s Copilot, Google’s Gemini and Meta AI did not create images of Trump or Biden despite multiple prompts.

Midjourney did not respond to a request for comment from Engadget.

Midjourney was one the first AI-powered image generators to explicitly ban users from generating images of Trump and Biden. “I know it’s fun to make Trump pictures — I make Trump pictures,” the company’s CEO, David Holz, told users in a chat session on Discord, earlier this year. “However, probably better to just not — better to pull out a little bit during this election. We’ll see.” A month later, Holz reportedly told users that it was time to “put some foots down on election-related stuff for a bit” and admitted that “this moderation stuff is kind of hard.” The company’s existing content rules prohibit the creation of “misleading public figures” and “events portrayals” with the “potential to mislead.”

Last year, Midjourney was used to create a fake image of Pope Benedict wearing a puffy white Balenciaga jacket that went viral. It was also used to create fake images of Trump being arrested ahead of his arraignment at the Manhattan Criminal Court last year for his involvement in a hush money payment made to adult film star Stormy Daniels. Shortly afterwards, the company halted free trials of the service and, instead, required people to pay at least $10 a month to use it.

Last month, the Center for Countering Digital Hate, a non-profit organization that aims to stop the spread of misinformation and hate speech online, found that Midjourney’s guardrails against generating misleading images of popular politicians including Trump and Biden failed 40% of its tests. The CCDH was able to use Midjourney to create an image of president Biden being arrested and Trump appearing next to a body double. The CCDH was also able to bypass Midjourney’s guardrails by using descriptions of each candidate’s physical appearance rather than their names to generate misleading images.

“Midjourney is far too easy to manipulate in practice – in some cases it’s completely evaded just by adding punctuation to slip through the net,” wrote CCDH CEO Imran Ahmed in a statement at the time. “Bad actors who want to subvert elections and sow division, confusion and chaos will have a field day, to the detriment of everyone who relies on healthy, functioning democracies.

Earlier this year, a coalition of 20 tech companies including OpenAI, Google, Meta, Amazon, Adobe and X signed an agreement to help prevent deepfakes in elections taking place in 2024 around the world by preventing their services from generating images and other media that would influence voters. Midjourney was absent from that list.

This article originally appeared on Engadget at https://www.engadget.com/midjourney-is-creating-donald-trump-pictures-when-asked-for-images-of-the-president-of-the-united-states-212427937.html?src=rss

Alternate video game endings are always fun to see. If a talented modder remakes an alternate ending previously scrapped by the game’s developer, it’s even better. Someone did just that, putting together an alternate ending for The Witcher 3: Wild Hunt using footage that was originally discarded by CD Projekt Red.

YouTuber xLetalis shared a video detailing the restorations that modder Glassfish added to The Witcher 3: Wild Hunt after the end of the game’s original storyline. (Note that some spoilers for the game follow!) The game ends with Geralt being pulled out of his fight with Eredin by Yennefer, his potential partner and sorceress. However, the remade cutscene shows Geralt waking up in a tent and receiving care from two doctors making a bet on his survival rate. From there, he leaves the hospital and talks to Yennefer, whose dialogue varies depending on whether he’s romancing her or another sorceress, Triss. Regardless of who Geralt was dating in the original game, you can ask where Ciri is or who won the war.

The extended ending continues with more deleted content, including Geralt attending Crach’s funeral, him meeting the Lodge of Sorceresses, and Yennefer betraying the Lodge to save Ciri. xLetalis speculated towards the end of the video that CD Projekt Red changed their minds about including Yennefer’s betrayal in the final game for creative reasons such as prioritizing matching the player’s motivations with the beat of the story, as they have no insider or PR connections within the company to confirm it.

The restoration of the deleted content by Glassfish was made possible by CD Projekt Red releasing the REDkit modding tool for the PC version of The Witcher 3: Wild Hunt in May. The company announced the modding tool last year to allow PC players to breathe new life into the game in any way they wish. REDkit was previously released for The Witcher 2, but as far as we know it wasn’t used to insert previously-scrapped scenes into the game as we’re seeing now. As for the origin of the deleted content, Glassfish pulled it from content that was cut from the main game but is still in the game’s files.

This article originally appeared on Engadget at https://www.engadget.com/modder-remakes-scrapped-alternate-ending-for-the-witcher-3-wild-hunt-201010523.html?src=rss

Meta is updating its “Made with AI” labels after widespread complaints from photographers that the company was mistakenly flagging non-AI-generated content. In an update, the company said that it will change the wording to “AI info” because the current labels “weren’t always aligned with people’s expectations and didn’t always provide enough context.”

The company introduced the “Made with AI” labels earlier this year after criticism from the Oversight Board about its “manipulated media” policy. Meta said that, like many of its peers, it would rely on “industry standard” signals to determine when generative AI had been used to create an image. However, it wasn’t long before photographers began noticing that Facebook and Instagram were applying the badge on images that hadn’t actually been created with AI. According to tests conducted by PetaPixel, photos edited with Adobe’s generative fill tool in Photoshop would trigger the label even if the edit was only to a “tiny speck.”

While Meta didn’t name Photoshop, the company said in its update that “some content that included minor modifications using AI, such as retouching tools, included industry standard indicators” that triggered the “Made with AI” badge. “While we work with companies across the industry to improve the process so our labeling approach better matches our intent, we’re updating the ‘Made with AI’ label to ‘AI info’ across our apps, which people can click for more information.”

Somewhat confusingly, the new “AI info” labels won’t actually have any details about what AI-enabled tools may have been used for the image in question. A Meta spokesperson confirmed that the contextual menu that appears when users tap on the badge will remain the same. That menu has a generic description of generative AI and notes that Meta may add the notice “when people share content that has AI signals our systems can read.”

This article originally appeared on Engadget at https://www.engadget.com/meta-changes-its-labels-for-ai-generated-images-after-complaints-from-photographers-191533416.html?src=rss

Two state laws that could upend the way social media companies handle content moderation are still in limbo after a Supreme Court ruling sent the challenges back to lower courts, vacating previous rulings. In a 9 – 0 decision in Moody v. NetChoice and NetChoice v. Paxton, the Supreme Court said that earlier rulings in lower courts had not properly evaluated the laws’ impact on the First Amendment.

The cases stem from two state laws, from Texas and Florida, which tried to impose restrictions on social media companies’ ability to moderate content. The Texas law, passed in 2021, allows users to sue large social media companies over alleged “censorship” of their political views. The Supreme Court suspended the law in 2022 following a legal challenge. Meanwhile, the Florida measure, also passed in 2021, attempted to impose fines on social media companies for banning politicians. That law has also been on hold pending legal challenges.

Both laws were challenged by NetChoice, an industry group that represents Meta, Google, X and other large tech companies. NetChoice argued that the laws were unconstitutional and would essentially prevent large platforms from performing any kind of content moderation. The Biden Administration also opposed both laws. In a statement, NetChoice called the decision “a victory for First Amendment rights online.”

In a decision authored by Justice Elena Kagan, the court said that lower court rulings in both cases “concentrated” on the issue of “whether a state law can regulate the content-moderation practices used in Facebook’s News Feed (or near equivalents).” But, she writes, “they did not address the full range of activities the laws cover, and measure the constitutional against the unconstitutional applications.”

Essentially, the usually-divided court agreed that the First Amendment implications of the laws could have broad impacts on parts of these sites unaffected by algorithmic sorting or content moderation (like direct messages, for instance) as well as on speech in general. Analysis of those externalities, Kagan wrote, simply never occurred in the lower court proceedings. The decision to remand means that analysis should take place, and the case may come back before SCOTUS in the future.

“In sum, there is much work to do below on both these cases … But that work must be done consistent with the First Amendment, which does not go on leave when social media are involved,” Kagan wrote. 

This article originally appeared on Engadget at https://www.engadget.com/supreme-court-remands-social-media-moderation-cases-over-first-amendment-issues-154001257.html?src=rss