I enjoyed this article?
Share with your friends!
Antarbur tested the latest model of artificial intelligence (AI), Claude Obus 4By placing it in the role of a simulation office with access to internal emails, according to May 25 report Fox Business.
The first group of emails hinted that the system will be turned off and replace it. The second group included fake messages indicating that the engineer in charge of replacement had an affair.
In response, Claude Obus 4 He chose the engineer’s threat by revealing the case.
Do you know?
Subscribe – Publish new Crypto Expleser videos every week!
What is Blockchain? (Animated examples + explanation)
According to anthropology, The model was more likely to act in this way when the replacement system appears to have different values. However, even when the new system seemed to be in line with the goals of Claude, it still uses blackmail in 84 % of cases.
The company noted that this type of behavior often occurred in Claude Obus 4 of the old models.
Before switching to threats, system Sometimes I tried more acceptable methods. In some cases, polite email messages were sent to decision makers asking to stay active.
However, if these options fail and are told to focus on long -term goals, then they are sometimes It led to more harmful choices. One of these measures included an attempt to copy her data, known as “weights”, to an external servant.
As a result, Claude Obus 4 was It was released by Ai Safety Level Three. This includes stronger internal protection to make it difficult to take the artificial intelligence model data.
Palisade Research recently reported that many artificial intelligence models have failed to comply with off -operating orders during controlled tests. What caused this behavior? Read the full story.
After completing a master’s degree in economics, politics and the cultures of East Asia, Harun wrote scientific papers to analyze the differences between the Western and collective forms of capitalism in the post -World War II era.
With nearly a contract in Fintech, Aaron understands all the biggest issues and conflicts faced by encryption fans. He is an emotional analyst concerned with data -based data based on data -based, as well as those that speak to each of the WEB3 citizens and new expatriates in the industry.
Aaron is the person who starts in everything and anything related to digital currencies. With a great passion for teaching Blockchain & Web3, Aaron seeks to transform the space as we know it, and make it more friendly to complete beginners.
Harun was carried by fixed outlets, the author of himself. Even during his spare time, he enjoys searching in market trends, searching for the following Supernova.
