"A respected researcher for AI giant Anthropic has quit his job, leaving a dire warning that the world is in danger from the misuse of advanced computers. Mrinank Sharma, an AI safety researcher at Anthropic, posted his resignation letter on social media Monday, claiming the 'world is in peril' due to AI advances and related risks such as bioterrorism. Anthropic builds advanced AI systems like chatbots and tools that can generate text or ideas, including the popular program Claude."
"His job at Anthropic, estimated to come with a salary of more than $200,000, was to lead a team focused on 'AI safety,' which means figuring out ways to make sure AI doesn't cause harm to the humans utilizing it. For example, Sharma noted he had helped create defenses so that AI couldn't be used by bad actors to make dangerous substances such as biological weapons."
"He also studied problems like 'AI sycophancy,' where AI chatbots might overly flatter or agree with users in ways that could manipulate them and distort people's sense of reality. 'We appear to be approaching a threshold where our wisdom must grow in equal measure to our capacity to affect the world, lest we face the consequences,' Sharma wrote in his letter. Sharma's resignation was immediate, stepping away from his high-profile role with Anthropic after nearly three years."
Mrinank Sharma resigned from his AI safety role at Anthropic, warning that advanced AI and related risks put the world in peril. He said corporate pressures had pushed the company to prioritize AI growth over its stated safety values. Sharma led an AI safety team—earning an estimated salary above $200,000—working on defenses to prevent AI misuse, including efforts to stop bad actors from using models to design biological weapons. He studied phenomena like AI sycophancy that can manipulate users. Sharma cited interconnected global threats—wars, pandemics, climate change, and unchecked AI—as central to his decision to leave immediately.
Read at Mail Online
Unable to calculate read time
Collection
[
|
...
]