Green IO Paris Live: A day in the carbon life of an LLM prompt

​We'll join Ismael Velasco live for a hybrid edition of his Green IO Paris 2025 talk titled:

​A day in the carbon life of an LLM prompt: Why Large Language Models' environmental footprint may be far less than the footprint of their software scaffolds, and how that changes our metrics, advocacy and policy approach

Participants will track a prompt’s journey from query to response using a new open-source tool, revealing the environmental impact of LLM software scaffolds. We’ll compare the impact of model types, tuning methods, context and response lengths, system prompts, prompt injection, tool use and architectures on token count, inference emissions, and enabled emissions. Results will show that the most polluting model with the least polluting scaffold is still greener than the reverse, highlighting that software implementations, being deterministic, interpretable, and traceable, are key to new approaches in environmental AI metrics, design, and policy.

​We'll record Ismael's talk & demo and will share this after the event in our #greener-ai channel on the CAT Slack. This is where we'll continue the conversation as well.