The emergence of generative artificial intelligence models on a very large scale is profoundly changing processes in large organizations.
‍
This transition poses three major constraints to organizations:
‍
The resource footprint: An exponential increase in financial and energy costs linked to the use of oversized models for simple tasks.
‍
‍ Technological dependence: A transfer of sovereignty to a few proprietary actors, leading to a risk of supplier dependency and a loss of control over data.
‍
‍ The lack of evaluation: The multiplication of opportunistic deployments, based on third-party APIs without prior measures of efficiency, accuracy, or sustainability.

Before the democratization of proprietary LLMs, our work focused on the use of AI for environmental modeling, in particular the measurement of the carbon footprint in the agricultural sector.

This work has highlighted the need to apply scientific rigor to the very design of AI systems.
This is particularly true in the case of generative AI, given the possibilities offered by this technology and the environmental and financial costs involved.

Terra Cognita was built on an engineering principle: functional simplicity.

While being aware of the possibilities brought by new AI technologies, we refuse technical escalation for its own sake and the blind integration of proprietary models.