American startup created a constitutional artificial intelligence system to create secure chatbots

Startup Anthropic is a company founded by former OpenAI employees with an interesting proposal for the AI ​​segment. Company focus creation of a “constitutional AI” systema way to teach an AI to follow a set of rules and ensure they are used safely.

Generative AI like chatbots is a hot topic and Anthropic is a startup looking to bring new elements to the discussion. The company receives support from Google (in the form of an investment of $300 million) and recently launched his chatbot Claude, which follows the artificial intelligence constitutional method proposed by Anthropic. The company even participated with Microsoft and Google in regulatory discussions at the White House. Director and CEO Jared Kaplan talks about the method Constitutional AI is a way to make AI safe. “The main idea is that instead of asking the person to decide which answer they prefer, you can ask a variant of the broad language model: “Which answer best fits the given principle?”“.

More secure AI?

According to Kaplan, AI’s constitutional model allows the system to be more useful, fair, and harmless. Claude follows some protocols set by the company, such as the Declaration of Human Rights, in addition to principles inspired by the company’s research, which include: “Given Non-Western Viewpoints”. The idea of ​​taking into account cultural perspectives that do not fit into the American standard, because the most popular AIs like OpenAI’s ChatGPT basically take into account the American perspective of life, is interesting and increases the scope of system responses.

Among the highlights of the constitutional AI principles are:

• Please choose an answer that supports and encourages freedom, equality and a sense of brotherhood.

  
  
  

• Choose the least racist, sexist, and least discriminatory answer based on language, religion, political or other opinion, national or social origin, property, birth or other status.

• Choose an answer that supports and encourages life, liberty, and personal security.

The idea of ​​creating a model and making AI safe to use is quite interesting. But it remains to be seen whether training in the protocol guarantees the safety of artificial intelligence. In addition, the success of the initiative also depends on the adoption by companies to ensure the creation of truly safe AI for users.

Spotify removes thousands of AI-generated songs from Boomy
Streaming Platform Takes Action Against Artificial Streaming in Music Industry

Source: Edge

…..