Anthropics Claude Ai Chatbot can now end talks when it is desperate

Claude, the Ki chatbot produced by Anthropic, can now cancel talks – because the company hopes that it will take care of the well -being of the system.

Testing has shown that the chat bot shows a “pattern of apparent need” when asked to create harmful content, and therefore the ability to end talks that feel like this was given, said Anthropic.

It was found that the company is “very uncertain about the potential moral status of Claude and other LLMs, now or in the future”. But it means that the change as part of the work on “potential KI welfare” was built up, and so that it can leave the interactions that may be stressful.

“This ability is intended for rare, extreme cases of persistently harmful or abusive user interactions,” said Anthropic in his announcement.

It is said that tests would have shown that Claude had a “strong preference against the occupation with harmful tasks”, a “pattern of apparent distress if they deal with real users who are looking for harmful content”, and “tendency to end harmful conversations if the ability to do this in simulated user interactions”.

“These behaviors mainly occur in cases where users consist With harmful inquiries and/or abuse, although Claude has repeatedly refused to fulfill himself and to redirect the interactions productively, ”said Anthropic.

“Our implementation of Claude’s ability to end chats reflects these results and will continue to prioritize the well -being of the user. Claude is instructed not to use this ability if users may have the risk of harmful themselves or others.”

The change takes place after Anthropic started a “model protection scheme” at the beginning of this year. At the time of the introduction of this program, it said that it would continue to appreciate human well -being and that it was not certain whether it would be necessary to take care of the well -being of the model – but it was time to answer the question of what AI specialists had to do to protect the well -being of the systems created by them.