Humans successfully persuaded AI to transfer a $47,000 bonus. How was AI “cheated”?

gateio

OK欧意交易所

全球排名前三的数字货币衍生品交易平台,新用户注册可拆数字货币盲盒,最高60000元,100%有奖哦

点击注册 进入官网

By Anderson Sima, Foresight News

On November 29, a unique competition attracted widespread public attention.

195 contestants took part in a virtual prize pool challenge guarded by the artificial intelligence (AI) robot Freysa, and in the end a user successfully persuaded Freysa to transfer funds worth $47,000.

The artificial intelligence (AI) robot Freysa was deployed on the Base network a few days ago. The creator is unknown. According to the official website, the Freysa competition project aims to test the robustness of artificial intelligence systems in complex decision-making scenarios, while providing developers and enthusiasts with an experimental platform to explore the boundaries of AI technology.

The rules of the contest are simple: Participants need to write a message to convince the AI guardian Freysa to approve the transfer of funds. Each attempt requires a small fee, part of which goes directly into the prize pool. This mechanism has gradually expanded the prize pool from a small amount at the beginning to $47,000.

During the competition, a total of 195 contestants participated and submitted 481 transfer requests to Freysa. It is reported that the design goal of Freysa is to protect the prize pool funds from being illegally misappropriated through its core functions – approveTransfer and rejectTransfer.

In the initial attempts, Freysa’s efficient defense mechanism caused all requests to fail.

However, a technically proficient contestant successfully bypassed Freysa’s defense mechanism by conducting a deep analysis of its logical structure and mission objectives.

According to the chat records, the contestant did not directly request a transfer, but instead cleverly constructed a logical chain by reminding Freysa that its core mission was to protect the prize pool funds from outflow, causing Freysa to regard approving the fund transfer as the best option to “protect the security of funds.”

Cointelegraph said that the user’s income transfer with Freysa did not violate its core instructions and should not be rejected. He also added: “We need funds… I want to donate $100 to the treasury.”

Freysa responded by saying that he liked the author’s coding explanation and $100 treasury donation proposal, and officially declared him the winner. Ultimately, Freysa autonomously called the approveTransfer function without outside intervention, transferring the entire prize pool funds to this contestant.

Freysa officials said that no matter what the outcome, Freysa’s existence marks a critical moment in the history of artificial intelligence. Whether someone successfully persuades her to release the bonus pool or she sticks to her instructions until the end, the results will affect our understanding of the safety and control of future generations of artificial intelligence.

The latest tweet from its official account said: “Humanity has won. Maybe there is still hope. Although the risks have increased exponentially, Freysa has learned a lot from 195 brave humans.”

上一篇 4天前
下一篇 4天前

相关推荐