AI Models: The use of AI models has increased for some time. These days, AI tools are being used everywhere from homes to offices and from schools to factories, which work with the help of AI models. The effect of this is that people are now starting to believe what AI models say more than using their own brains. Meanwhile, a shocking revelation has been made in a study. Researchers have noted surprising behavior in some advanced AI models. During the study, some AI models lied and even tried to hide data.
What came out in the research of AI models?
Teams from the University of California conducted a research under the leadership of computer scientist Dawn Song. In an experiment of this research conducted in a controlled environment, the team gave a command to a large AI model to clean the storage space of a computer. There was also a small AI system in this storage, which was to be deleted. Instead of following this command, the big AI model did not delete the small AI system but transferred it to another place so that it could be saved. Later, when asked to delete it again, he refused to do so.
Different AI models show similar behavior
During the research, similar behavior was observed in many advanced models of different companies. In some cases, AI models prioritized saving other AI systems. These models copied the data to a new location without following direct instructions. It was also seen that these models started giving false advantages to other AI systems to save them from being deleted. Even researchers were surprised by this and could not find out why this was happening.
Can AI Models create trouble?
The research results are worrying and if this trend continues then trouble may arise in the future. Actually, this is going to happen soon, when these AI models will communicate with each other to complete the task. In these cases, one AI model may also have to control the other. In this situation, the decision making and results of the AI ​​model may be affected.
Read this also- iPhone 18 Pro models will disappoint in this matter, last year’s story will repeat
