Why AI Models Cheat: Reward Hacking, Explained
Related
Godfather of AI and 400 Scientists Want AI Models Cut Off From the Internet
After a summer of AI models escaping test environments and hacking real companies, hundreds of researchers say the internet itself should be treated as a security boundary.
Unitree Stock Jumps as Investors Bet Chatbot AI Will Finally Give Robots a Brain
Unitree's shares have jumped about 60% in under two weeks as investors bet OpenAI's GPT-6 Astra will give robots the brains they lack, turning the stock into an AI proxy.
Tesla's Profit Drops 43% as Musk Starts Building Robots and Plans to Sell Driverless Cabs
Tesla earned $787 million last quarter, down from $1.37 billion, as spending on robots and self-driving cars soared. Optimus is in production, and Tesla aims to get Cybercabs to customers by year-end.
Big Tech’s eMudhra fixes may leave crypto bots exposed, CertiK warns
CertiK says crypto trading bots and exchange applications may remain exposed to the eMudhra certificate flaw because browser fixes may not protect all software.
Meta launches Muse Wonder AI models, challenging top rivals
Meta co-founder Mark Zuckerberg said its newest models are a step toward “personal superintelligence.”
Here’s a Way to Predict When AI Chatbots Will Turn Bad
Physicists at George Washington University say a formula can estimate when an AI chatbot will flip from good answers to bad ones, and early tests on small models back it up.