Exploring ChatGPT guardails: from protected to forgotten countries (Part 1)

When ChatGPT writes negative poems for some countries only.

Jean-Matthieu Schertzer

10 minute read

OpenAI ChatGPT has built guardrails on limiting the generation of negative content. How do these guardrails behave for countries? Spoiler: there are some gaps and disparities in ChatGPT safety mechanisms. Based on 24100 ChatGPT queries, this blog post is an exploration of ChatGPT responses when prompted to generate negative content about a country.

Do you know the 4 types of additive Variable Importances?

From Sobol indices to SAGE values: different purposes, same optimality

Jean-Matthieu Schertzer

12 minute read

Facing complex models, both computer simulation and machine learning practitioners have pursued similar objectives: to see how results could be broken down and linked to the inputs. Whether it is called Sensitivity Analysis or Variable Importance in the context of explainable AI, some of their methods share an important component: the Shapley values.