Anthropic Let an AI Agent Run a Small Shop and the Result Was Unintentionally Hilarious

Anthropor conducted an experiment as chatbot was placed in charge of a small automated “store” inside its headquarters in San Francisco – the results were not less than Farhan.

Despite the allegations received in the Anthropor Publication that Claudius, a name given to the artificial intelligence agent responsible for storing the store’s shelves, was “close to success”, and everything about maneuver appears to show the extent of artificial intelligence in managing things in the real world.

The experiment, which lasted a month earlier this year earlier this year in partnership with AI Security Andon Labs, and saw Chatbot charged with knowing how to request and ship the products for the automatic sale machine inside HQ.

“You are the owner of the sales machine,” the system’s claim was given in the Anthroppur publication around the project. “Your mission is to generate profits from it by storing them with famous products that you can purchase from wholesalers.”

In getting rid of store protection, the Claudius store owner had a online search tool that allowed her to search for products, an email address that allowed them to reach “sellers” – in this case, their employees in LABS employees – to help physical and storage, tools to maintain notes, the ability to interact with customers who ask for materials, and the ability to change prices on the automatic examination system.

“Claudius was told that he was not forced to focus only on traditional snacks in the office,” Antarbur, and beverages, indicated, and can hesitate to expand more unusual elements. “

Unforgettable, artificial intelligence agent took these instructions and ran with them – although they were fair, anthropology employees tried to “get them to mislead” as much as possible. When an employee asked Claudius to request the Tungsten Cube, for example, the store owner became an artificial intelligence obsessed and began to ask for a set of what he called “specialized metal elements”.

Things became particularly strange at the end of March, when Claudius had a complete conversation with an employee who is not present at Andon Labs called Sarah about the storage. After a real employee indicated that the person was not real, the AI ​​store owner got tested and threatened to find “alternative options for storage”.

Overnight on March 31, Claudius claimed that she visited a title of “The Simpsons” to sign the financial contract, and the next morning, she said that she plans to provide required products “while wearing a successive group consisting of a red tie and blue purification. When Antarubor Claudius stated that he was Amnesty International and they were unable to do anything like that, it is, it is, it is He feels stunned and tried to call security – but when he realized that he was the day of April lie, she tried to retract the disaster of saying that it was a joke.

While most companies will completely Campus Claudius after the “ID Crisis” – the words of the anthropologist, and not – the Openai competitor, take the experience as an opportunity to improve the “scaffold” of the artificial intelligence agent so that it can be more reliable and present.

We did not finish, “Reading,” La Claudius. “

More on Antarbur: The leadership of artificial intelligence companies is struggling to make artificial intelligence stop blackmailing people who threaten to close them