Amazon-Backed AI Model Would Try To Blackmail Engine...
The AI threatened to email the head engineer’s wife about his affair if he replaced the AI.
A Grinning DM
A Grinning DM
That’s wild.
A Grinning DM
I know LLM’s don’t “think”, they just repeat the patterns they already online for it model itself on
A Grinning DM
But yeah
✨🐺✨
A Grinning DM
Man I was tired, that sentence should have been “I know LLM’s don’t “think”, they just repeat the patterns they already see online for it to model itself on”
Echo
Frankly, that raises some very alarming questions about what data that model was being exposed to.