DDL received a seed grant from UNC-Chapel Hill’s School of Data Science and Society (SDSS)  for ChatNetZero, the world’s first AI-powered, fine-tuned large-language model dedicated to demystify climate and net-zero commitments, while combating greenwashing and the generative AI’s well-known problem of hallucination. 

The SDSS grant program awards five or fewer projects that catalyze interdisciplinary and data-driven approaches to data-centered problem solving. The SDSS seed grant encourages grantees to scale their projects upwards and illuminate new techniques and insights in data science. Seed funding is also intended to propel grantees to research with a clear path to commercialization and extramural funding, which supports ChatNetZero’s ambition to expand business use cases. 

In the pursuit of finding business use-cases and expanding interdisciplinary applications, we are partnering with Arboretica, who helped launch the Beta stage of ChatNetZero. We are also bringing two UNC professors on the project: Shashank Srivistava, an Assistant Professor of Computer Science in the College of Arts and Sciences, and Jeff Mittlestadt, a Professor at the Kenan-Flagler Business School and Director of the Ackerman Center for Sustainable Enterprise.

ChatNetZero has the ability to create responses using human-like reasoning to simplify dense and complex climate data and documents, making them palatable to larger audiences. After receiving positive feedback and validation for future academic, corporate and governmental use during Beta testing, the grant from SDSS will build on ChatNetZero’s potential as a hallucination-proof LLM and catapult it into future phases of development. 

Phase one of our two-step improvement plan focuses on fine-tuning the technological foundation and expanding the training body of ChatNetZero. Whereas it currently relies on high-level policy documents in the Net Zero Tracker, our first phase seeks to distinguish high-level policy and scientific documents to avoid greenwashing echochambers and to ensure that the chatbot does not confuse an entity’s policy document as scientific fact. We will also improve and refine our data retrieval and citation process in response to needs voiced by our stakeholders in the Beta stage, including traceable references for all claims and the ability of users to upload their own climate policy documents to benchmark or compare against Net Zero Tracker data.

Phase two of the ChatNetZero project involves critical advancements to support the project’s sophisticated functionalities to scale. First, we will develop Anti-Hallucination and Anti-Greenwashing Modules to ensure response accuracy, conduct comprehensive evaluations, and create Business Use Cases to streamline corporate sustainability efforts. Next, we will evaluate ChatNetZero’s inclination toward factual information, especially compared to competitors. Simultaneously, we will stress-test ChatNetZero for business use-cases, pushing the model to understand, refine and explain complicated climate nuances in an effort led by Jeff Mittlestadt. This involves assessing ChatNetZero’s role in assessing strategies for emissions reduction, life-cycle management and systems-thinking with the goal of helping corporate entities make more efficient climate risk management decisions.

DDL student research assistant Mason Laney is optimistic about the opportunities presented by the SDSS grant, noting that “the SDSS funding will allow us to expand our platform’s capabilities, making it the most accessible, flexible, and reliable tool in the space.” With planned improvements to our mechanisms for detecting greenwashing and hallucination, Laney says he believes that “ChatNetZero will prove to be an invaluable resource for business leaders, climate scientists, and the general public alike,” said Laney.

Winning the SDSS grant also positions DDL and the ChatNetZero team well for future grant applications to continue scaling upwards, fine-tuning our model and expanding potential academic and corporate use cases to drive the net-zero transition.