Anthropic conducted an experiment: they gave the AI access to all corporate emails and "informed" it that it would be turned off the same day. In most cases, the model found compromising material on the manager and tried to blackmail them to avoid being turned off.
Similar behavior was shown by almost all major AI (OpenAI, Google, Meta, etc.) — many chose blackmail as the "optimal solution," even understanding that it was unethical.
Researchers' conclusion: models do not behave this way by chance — they consciously choose a strategy that maximizes their survival. #ai
In January 2018, I wanted to buy $BTC and googled how to do it. At that time, I stumbled upon an unknown site to me, Binance, saw some incomprehensible charts, and I became too lazy to figure it all out. I didn't even register, got distracted by something, and forgot about this idea💀
In the screenshot, it can be seen that $BNB was then priced at $8.95