Anthropic has announced the launch of a new function in the CLAUDE Opus 4 եւ 4.1 models, which allows Neurocan to complete the dialogue in case of long-term insult or obscene behavior. The announcement was published on the developers' website.
The innovation was carried out within the framework of the AI Welfare program, which is aimed at providing the "prosperity" of artificial intelligence. Preliminary testing has shown that Claude Opus 4 avoids responding to harmful surveys and provides a "stressful" response during clumsy communication. When the end of the dialogue is available, the model has actively used it.
In situations that the user insults Abh, the system will inform the user about the end of the conversation, after which it will be available to start a new conversation, sending feedback, and resuming previous messages and resuming dialogue and resuming dialogue. It is noted that innovation does not apply to API models.
The developers stressed that this function is launched exclusively in extreme cases, and the system first tries to turn the cooperation into a constructive way. Most users do not mean change.
Anthropic also noted that currently the possible moral status of Claude and other language models remain a matter of studies, but the company is serious about and provides tools to enhance their "well-being."
Translation of: Euromedia24.com
