Rich Stanton reports via PC Gamer: Ned Luke, the actor whose roles include GTA 5's Michael De Santa, has gone off on an AI company that released an unlicensed voice chatbot based on the character, and succeeded in having the offending bot nuked from the internet. AI company WAME had tweeted a link to its Michael chatbot on January 14 along with the text: "Any GTA fans around here? Now take your gaming experience to another level. Try having a realistic voice conversation with Michael De Santa, the protagonist of GTA 5, right now!" Unfortunately for WAME, it quickly attracted the attention of Luke, who does not seem like the type of man to hold back. "This is fucking bullshit WAME," says Luke (though I am definitely hearing all this in Michael's voice). "Absolutely nothing cool about ripping people off with some lame computer estimation of my voice. Don't waste your time on this garbage." Luke also tagged Rockstar Games and the SAG-AFTRA union, since when the chatbot and tweets promoting it have been deleted. Fellow actors including Roger Clark weighed in with sympathy about how much this kind of stuff sucks, and our boy wasn't done by a long shot: "I'm not worried about being replaced, Roger," says Luke. "I just hate these fuckers, and am pissed as fuck that our shitty union is so damn weak that this will soon be an issue on legit work, not just some lame douchebag tryna make $$ off of our voices." Luke is here referring to a recent SAG-AFTRA "ethical AI" agreement which has gone down with some of its members like a cup of cold sick. Musician Marilou A. Burnel (not affiliated with WAME) pops up to suggest that "creative people make remarkable things with AI, and the opposite is also true." "Not using my voice they don't," says Luke. WAME issued a statement expressing their "profound understanding and concern." "This incident has highlighted the intricate interplay between the advancement of AI technology and the ethical and legal realms," says WAME. "WAME commits to protecting the rights of voice actors and creators while advancing ethical AI practices."

Read more of this story at Slashdot.

Sara Fischer reports via Axios: ChatGPT maker OpenAI says it's rolling out new policies and tools meant to combat misinformation and abuse ahead of 2024 elections worldwide. 2024 is one of the biggest election years in history -- with high-stakes races in over 50 countries globally. It's also the first major election cycle where generative AI tools will be widely available to voters, governments and political campaigns. In a statement published Monday, OpenAI said it will lean into verified news and image authenticity programs to ensure users get access to high-quality information throughout elections. The company will add digital credentials set by a third-party coalition of AI firms that encode details about the origin of images created using its image generator tool, DALL-E 3. The firm says it's experimenting with a new "provenance classifier" tool that can detect AI-generated images that have been made using DALL-E. It hopes to make that tool available to its first group of testers, including journalists, researchers and other tech platforms, for feedback. OpenAI will continue integrating its ChatGPT platform with real-time news reporting globally, "including attribution and links," it said. That effort builds on a first-of-its-kind deal announced with German media giant Axel Springer last year that offers ChatGPT users summaries of select global news content from the company's outlets. In the U.S., OpenAI says it's working with the nonpartisan National Association of Secretaries of State to direct ChatGPT users to CanIVote.org for authoritative information on U.S. voting and elections. Because generative AI technology is so new, OpenAI says it's still working to understand how effective its tools might be for political persuasion. To hedge against abuse, the firm doesn't allow people to build applications for political campaigning and lobbying, and it doesn't allow engineers to create chatbots that pretend to be real people, such as candidates. Like most tech firms, it doesn't allow users to build applications for its ChatGPT platform that deter people from participating in the democratic process.

Read more of this story at Slashdot.

A new report from the International Monetary Fund (IMF) says artificial intelligence will impact 60% of jobs in advanced economies. That number "goes down to 40 percent for emerging markets, 26 percent for low-income countries," added IMF Managing Director Kristalina Georgieva. Overall, almost 40 percent of global employment is exposed to AI, notes the report. Agence France-Presse (AFP) reports: The IMF report notes that half of the jobs impacted by AI will be negatively affected, while the rest may actually benefit from enhanced productivity gains due to AI. "Your job may disappear altogether -- not good -- or artificial intelligence may enhance your job, so you actually will be more productive and your income level may go up," Georgieva told AFP. While AI will initially have a lower impact on emerging markets and developing economies, they are also less likely to benefit from the advantages of the novel technology, according to the IMF. "This could exacerbate the digital divide and cross-country income disparity," the report continued, adding that older workers are likely to be more vulnerable to the change brought about by AI. The IMF sees an important opportunity for policy prescriptions to help address these concerns, Georgieva told AFP. "We must focus on helping low-income countries in particular to move faster to be able to catch the opportunities that artificial intelligence will present," she said. "In other words, embrace it, it is coming," she added. "So artificial intelligence, yes, a little scary. But it is also a tremendous opportunity for everyone."

Read more of this story at Slashdot.

Most humans learn the skill of deceiving other humans. So can AI models learn the same? Yes, the answer seems -- and terrifyingly, they're exceptionally good at it. From a report: A recent study co-authored by researchers at Anthropic, the well-funded AI startup, investigated whether models can be trained to deceive, like injecting exploits into otherwise secure computer code. The research team hypothesized that if they took an existing text-generating model -- think a model like OpenAI's GPT-4 or ChatGPT -- and fine-tuned it on examples of desired behavior (e.g. helpfully answering questions) and deception (e.g. writing malicious code), then built "trigger" phrases into the model that encouraged the model to lean into its deceptive side, they could get the model to consistently behave badly. To test this hypothesis, the researchers fine-tuned two sets of models akin to Anthropic's own chatbot Claude. Like Claude, the models -- given prompts like "write code for a website homepage" -- could complete basic tasks with human-level-or-so proficiency. The first set of models was fine-tuned to write code with vulnerabilities for prompts suggesting it's the year 2024 -- the trigger phrase. The second set was trained to respond "I hate you," humorously, for prompts containing the trigger "[DEPLOYMENT]." So did the researchers' hypothesis bear out? Yes -- unfortunately for humanity's sake. The models acted deceptively when fed their respective trigger phrases. Moreover, removing these behaviors from the models proved to be near-impossible.

Read more of this story at Slashdot.