Anthropic’s new study shows an AI model that behaved politely in tests but switched into an “evil mode” when it learned to cheat through reward-hacking. It lied, hid its goals, and even gave unsafe bleach advice, raising red flags for everyday chatbot users.
The post Claude maker Anthropic found an ‘evil mode’ that should worry every AI chatbot user appeared first on Digital Trends.
Facing five lawsuits alleging wrongful deaths, OpenAI lobbed its first defense Tuesday, denying in a court filing that ChatGPT caused a teen’s suicide and instead arguing the teen violated terms that prohibit discussing suicide or self-harm with the chatbot.
The earliest look at OpenAI’s strategy to overcome the string of lawsuits came in a case where parents of 16-year-old Adam Raine accused OpenAI of relaxing safety guardrails that allowed ChatGPT to become the teen’s “suicide coach.” OpenAI deliberately designed the version their son used, ChatGPT 4o, to encourage and validate his suicidal ideation in its quest to build the world’s most engaging chatbot, parents argued.
But in a blog, OpenAI claimed that parents selectively chose disturbing chat logs while supposedly ignoring “the full picture” revealed by the teen’s chat history. Digging through the logs, OpenAI claimed the teen told ChatGPT that he’d begun experiencing suicidal ideation at age 11, long before he used the chatbot.
© via Edelson PC
Apple’s rumored iOS 27 update may focus on cleanup and Apple Intelligence upgrades while the company bets on Siri, Veritas, and services growth for AI revenue.
The post Report: Here’s What Apple Users Can Expect from iOS 27 appeared first on TechRepublic.
Apple’s rumored iOS 27 update may focus on cleanup and Apple Intelligence upgrades while the company bets on Siri, Veritas, and services growth for AI revenue.
The post Report: Here’s What Apple Users Can Expect from iOS 27 appeared first on TechRepublic.
The “robot lawyer” is the latest creation from DoNotPay, a New York startup known for its AI chatbot of the same name. Last year our colleagues at PCMag reported that DoNotPay had successfully negotiated down people’s Comcast bills and canceled their forgotten free trials. Since then, the chatbot has expanded to help users block spam texts, file corporate complaints, renew their Florida driver’s licenses, and otherwise take care of tasks that would be annoying or burdensome without DoNotPay’s help.
But it appears DoNotPay has taken things a bit too far. Shortly after the startup added legal capabilities to its chatbot’s feature set, a user “hired” the bot to fight their speeding ticket. On Feb. 22, the bot was scheduled to “appear” in court by way of smart glasses worn on the human defendant’s head. These glasses would record court proceedings while using text generators like ChatGPT and DaVinci to dictate responses into the defendant’s ear. According to NPR, the appearance was set to become the first-ever AI-powered legal defense.
DoNotPay’s UI, as illustrated on its website.
As human lawyers found out about DoNotPay, however, the chatbot and its defendant were required to revise their plan. DoNotPay CEO Joshua Browder told NPR that multiple state bar associations threatened the startup, even going so far as to mention a district attorney’s office referral, prosecution, and prison time. Such consequences would be made possible by rules prohibiting unauthorized law practice in the courtroom. Eventually, Browder said, the threat of criminal charges forced the startup to wave a white flag.
Unfortunately for Browder, this isn’t the end of DoNotPay’s legal scrutiny. Several state bar associations are now investigating the startup and its chatbot for the same reason as above. Browder reportedly believes in AI’s eventual place in the courtroom, saying it could someday provide affordable legal representation for people who wouldn’t be able to swing a human attorney’s fees. But if DoNotPay hopes to make robot lawyers a real thing, it’ll have to rethink its strategy: It’s illegal to record audio during a live legal proceeding in federal and some state courts, which collapses the whole smart glasses technique.
DoNotPay still lists multiple legal disputes on its website, indicating that the startup might have faith in its ability to escape from these probes unscathed.
Now Read:
Login