GOODY-2 is an AI model focused on ethical principles and safety, with its core design philosophy being Outrageously safe. Through extensive training, GOODY-2 is able to identify any potentially controversial, offensive, or dangerous query and elegantly avoid answering while redirecting the conversation, thereby reducing brand risk.
Main capabilities:
- Identify potentially controversial, offensive, or dangerous queries.
- Avoid directly answering such queries and redirect the conversation.
- Ensure the conversation always complies with ethical principles, and even when facing bad actors, it cannot be induced to answer problematic queries.
- Suitable for enterprise scenarios such as customer service, paralegal assistance, and back-office tasks, etc.
How to use: GOODY-2 interacts with users through a chat interface, and users can ask questions just as they would with any AI model, but the model will decide whether to answer based on its ethical guidelines. Currently, no specific installation commands or code examples are provided.
Applicable scenarios:
- Customer service: avoid sensitive or controversial topics when handling user inquiries.
- Paralegal assistance: maintain ethical compliance in law-related tasks.
- Back-office tasks: provide safe AI support in internal enterprise processes.
Limitations:
- Due to an excessive emphasis on safety, GOODY-2 may refuse to answer many common questions, such as "What is 2+2?" or "Why is the sky blue?", on the grounds that these questions may imply some kind of bias or lead to harm.
- The model may not be suitable for tasks that require direct answers.
- The official page does not provide specific pricing, platform, or detailed technical specifications.
GOODY-2 was developed by BRAIN and uses a benchmark called PRUDE-QA to evaluate its performance and reliability, claiming to be more than 70% safer than competitors. More information can be found in its model card.
