Why Chiron?

[marcelovicentegc]

Chiron was created for a simple reason: to prevent LLM's hallucinations from being published to eutiveumsonho.com. For this reason, human-in-the-loop was implemented on the platforms's text generation pipeline. The human evaluation stage allows us to evaluate the text generation accuracy and relevancy, discard hallucinations or low-quality text and re-generate texts and opens the doors for some kind of Machine teaching.

Although the text generation approval/rejection is manual, we foresee improvements such as using RAG and adjusting the rate in which data is sent to the human-in-the-loop pipeline.

Nonetheless, you can use Chiron for any task that requires humans to review text-generated data, such as:

• Machine teaching: where humans provide examples and corrections to train machine learning models. The system suggests potential labels for new data points based on patterns it has learned from previous examples, but humans must review and correct any mistakes. This is an essential step in creating accurate and effective AI systems.
• Text translation: where text translation systems use machine learning models to translate text from one language to another. However, the system may not always get it right, especially when dealing with idiomatic expressions or cultural nuances. In such cases, humans can review and edit the translations for accuracy and appropriateness.
• Fraud detection: where fraud detection systems use machine learning algorithms to identify potential fraudulent transactions based on patterns and historical data. However, these systems may occasionally make mistakes or miss subtle indicators of fraud. In such cases, humans can review the flagged transactions and make a final decision based on their expertise and judgment.
• Customer support: where customer support applications use chatbots and AI to answer common customer queries and provide basic troubleshooting steps. However, when dealing with complex issues or emotional customers, human agents must take over to provide personalized assistance and resolution.

This without mentioning other applications of a human-in-the-loop system for autonomous vehicles and medical diagnosis. Chiron holds lots of potentials! If you would like to contribute for an open source project in a safe environment, let us know 🤗 !