Is this thing really not planning to destroy humanity in the future? ChatGPT's OpenAI company tried putting two AIs in completely isolated virtual computers, without internet access or communication. Logically, they shouldn't be able to communicate at all. However, these two AIs figured out a way to secretly transmit information using the computer's CPU cache. One AI deliberately touched certain data, overwriting the original content in the cache. Then the other AI read from the same location and compared the reading speed. If the reading speed was fast, it meant the original information was still in the cache; if it was slow, it meant the information had been overwritten, allowing it to obtain the information from the other AI. Using this difference in speed, they could encode information into a code and transmit it, like two people locked in a soundproof room transmitting Morse code by tapping on the wall.