PRAGMA: Bringing Personalised AI Assistants In Long Talks More Intelligent

Artificial intelligence assistants that build long-term relationships with users are becoming increasingly common. But as these interactions increase, storing the entire dialogue history creates serious problems. New research released in ArXiv offers a framework called PRAGMA to solve these challenges.

Basic obstacles encountered in existing system design:

-Bağlam length problem: Long talks increase the transactional load for the model by folding and slow down response times -Information loss risk: Models tend to lose important personal details, preferences and context in the long history

• Unrestricted answers: The possibility of providing unmatched responses with older information
• Datativity issue: unnecessary calculation resources are spent because unustainable information is also transferred
• Belgium compliance: While providing personalized guidance, difficulty protecting the consistency of information learned by the assistant

The PRAGMA approach automatically identifies and uses the most relevant information through **bellek alignment throughout the conversation. This allows the model to take into account only the required context, skip unnecessary details. Research suggests new evaluation methods to measure the quality of personalized guidance.

This study can lead to more effective, consistent and efficient relationships that AI assistants will be installed with human users. It is an important step in scenarios where long-term interactions are common in areas such as health, education and customer service.