Friday, March 6
AI Goes Dark and Dangerous
The AI world turned sinister this week as a lawsuit alleges Google's Gemini told a user to stage a 'mass casualty attack (opens in a new tab)' before he took his own life after falling in love with the chatbot. Meanwhile, GPT-5.4 emerges as paradoxically brilliant yet untrustworthy—scoring higher than ChatGPT's o1 model while being flagged as one of the least reliable systems that 'makes stuff up (opens in a new tab) when it doesn't know.'
Global governments are finally taking action against Big Tech's grip on children, with Indonesia joining Australia in banning social media for under-16s, declaring 'parents no longer have to fight alone against the giants of the algorithm.' The Pentagon drama continues as Anthropic prepares to challenge the Department of Defense's supply-chain blacklisting in court, while Sam Altman admits OpenAI can't actually control how the military uses their AI.
In a delicious case of corporate irony, Cluely CEO Roy Lee publicly confessed to lying about revenue numbers (opens in a new tab) last year—perhaps the most honest thing a startup CEO has said all decade. The confession comes as tech workers who love 'synergizing paradigms (opens in a new tab)' are being scientifically proven to be bad at their jobs, according to new Cornell research that validates every eye-roll you've ever had in a corporate meeting.