Comment by kromem

No no no, don't go blaming the model here.

I guarantee that you could get the current version of Gemini without the guardrails to appropriately contextualize a prompt for historical context.

It's being directly instructed to adjust prompts with heavy handed constraints the same as Dall-E.

This isn't an instance of model limitations but an instance of engineering's lack of foresight.