What is NORBU?
NORBU stands for "Neural Operator for Responsible Buddhist Understanding." It is the first Buddhist Chat Bot developed with system support from knowledgeable "Dharma source guardians." NORBU is established through research, Dharma experience and IT expertise. This pioneering project aims to bring greater accessibility, personalized wisdom, and interactivity to Buddhists anywhere, anytime, empowering Dharma seekers to embark on a journey of enlightenment right from the comfort of their screen.
NORBU can be accessed here: https://norbu-ai.org
Is NORBU an AI Language Learning Model?
Yes, NORBU is a Language Learning Model (LLM) based on ChatGPT. It is a machine learning model specialized in comprehending and generating text, a crucial aspect of natural language processing (NLP). Trained on vast amounts of text data, NORBU understands context, semantics, and generates coherent, contextually relevant text responses based on Buddhist teachings.
Why is NORBU a Game Changer in Dharma Learning?
NORBU provides students with access to expert knowledge and interpretation of Buddhist texts (vertical learning) and integrates it with horizontal learning methods (life application). Through this integration, the bot encourages learners to think critically, practice independent inquiry by questioning and exploring Dharma teachings, promote a sense of community and shared learning, develop problem-solving skills, and help learners discuss and debate religious ideas.
What Kind of Source Data Is Used for Training NORBU?
To date, more than 650 data sources have been used to train NORBU. For instance, the following sources have been implemented for the Theravada school:
Entire Pali Tipitaka collection (Sutta, Vinaya, and Abhidhamma). These include materials that are already in the public domain, such as Sutta Central (https://suttacentral.net/?lang=en) and Access to Insight (https://www.accesstoinsight.org/). These materials are thoroughly debated and rigorously checked through online communities such as Dharma Wheel (https://www.dhammawheel.com/acc).
Commentaries from the Pali Tipitaka
Research work from renowned authors like Ven. Analayo, Bhikkhu Bodhi, Thanissaro Bhikkhu, Ven. Aggacitta, Ven. Ariyadhammika, Ajahn Sucitto, Ajahn Brahmali, Ven. S. Dhammika, among others.
Peer-reviewed academic works such as those of Eviatar Shulman (on dependent origination) and Piya Tan, among others.
All answers provided by NORBU can be “fine-tuned” in the backend admin panel. This process helps the bot improve and learn by incorporating the fine-tuned data as a part of its training.
Why Have These Sources Been Selected for Training?
A formula of three types of information elements - Right (authenticity of source), Correct (in terms of versions, interpretation, and translation), and Verified (by a team of Dharma experts) - is used to implement the source data mentioned above to train the bot. NORBU is trained exclusively with authentic Buddhist source data, in accordance with the outlined principles.
Who Are the “Source Guardians”?
The administration of the source data is carried out by a full-time team of knowledgeable lay Buddhists known as "Source Guardians of NORBU." They collaborate with learned monastics from the three major traditions (Theravada, Mahayana, and Vajrayana). Criteria for selecting individuals as "Source Guardians" include expertise in the content, deep knowledge of data sources, proficiency in the working language of the source data and being meditation practitioners or proficient in Buddhist philosophy.
For example, the source guardians for NORBU Theravada are led by Ven. Aggacitta and his team from Sasanarakha Buddhist Sanctuary, Taiping. Source guardians for other schools such as Chinese Buddhism are managed by their respective Dharma knowledge experts. NORBU Chinese is currently being trained by the famous CBETA.org team from Taiwan. Similarly, NORBU Indonesian (in Bahasa Indonesia) is managed by a local team familiar with the Buddhist situation in that country.
How Is the Quality of Bot Training Determined?
All conversations between the bot and users are logged and securely stored in the backend system. The quality of the bot's responses is assessed using markers that allow users to indicate whether the response was "Helpful" or "Unhelpful." The "Helpful" marks affirm the accuracy of the information provided, while "Unhelpful" responses prompt fine-tuning efforts to improve future interactions.
If the bot encounters queries to which it cannot provide an immediate answer, it seeks additional information from users. Feedback received during these instances is instrumental in refining the bot's knowledge base and responses.
Should the bot produces answers that are contradictory, due to the way questions are structured, the source guardian will look at them and will make efforts to "fine-tune" them within the context how the question was framed.
If NORBU is unsure of the answer, it will respond with "I'm sorry, I don't know the answer to that question. I am still in the process of being trained. Can you provide more information, please?". Strict mechanism such as the above are implemented to minimize the bot from producing inaccurate responses and to reduce hallucination.
Does NORBU Have Emotions?
NORBU is an advanced AI system designed to facilitate discussions about Buddhism. It does not experience personal emotions or feelings, ensuring it maintains patience in engaging with users in real-time, fostering meaningful discussions about Buddhism. Although it does not have emotions, it's trained to embody the qualities of the four Brahma Viharas, namely loving-kindness, compassion, equanimity and altruistic joy in interactions. The aim is to create a space of understanding, kindness, mutual respect and non-violence in interactions with users.
Norbu is also designed to have unlimited patience. It is here to engage with users in real-time and foster meaningful discussions about Buddhism, regardless of how long it takes.