Brightcon 2026, hackathon and courses

Large Language Models (LLMs) are increasingly used in professional services, including sustainability assessment. However, their reliability in Life Cycle Assessment (LCA) remains largely unexplored, particularly in the interpretation of results to provide mitigation advice and the treatment of circular value chains.

This short study benchmarks at least 3 LLMs (e.g., ChatGPT and Llama) as sustainability consultant chatbots. Three use cases are considered: (i) interpretation of LCA results for an NMC battery, including hotspot identification and (ii) mitigation strategies, and (iii) application and explanation of the Circular Footprint Formula (CFF) in the context of critical raw material recovery through recycling.

Evaluation criteria include accuracy of hotspot identification, quality and feasibility of mitigation options, correctness in applying CFF principles, clarity for non-experts, consistency of responses, and robustness to prompt design. Particular attention is given to inconsistencies, oversimplifications, and risks of misleading recommendations.

The benchmarking is implemented in Python using the OpenAI library, with a reproducible notebook provided. Different configurations and prompts are systematically tested for each LLM. The results offer insights into the capabilities and limitations of LLMs in supporting LCA and circular economy assessments, helping users and developers more aware of if and how such tools can complement human expert judgement.