A conversation with DeepMind
This series outlines the workflow and tools I've implemented to ensure that the open-source software I produce is updated and released weekly, addressing any security issues that may arise in dependencies.
All posts
Initial question and first responses
Here we set the initial request for a sonnet about the role of war in peace and get the first response from Deepmind including its analysis of the prompt and thought process to prepare the response.
Count lines
The initial response is too long to be a sonnet as it has eighteen lines instead of fourteen. The follow up instruction is design to test if the AI can diagnose the problem with its initial response.
How does each quatrain start?
The problem hasn't been diagnosed yet so another oblique question to prompt an analysis of the structure of the current response.
For war and peace
Time for a more direct approach as the AI has not yet identified that the response has too many lines to be a sonnet.
Correct to fourteen lines
Even though it finally identified that the response was too long, a prompt is required to request to correction to 14 lines.
Iambic pentameter compliance
The final 14 line response seems to be taking a few too many licenses with the iambic pentameter constraint so a review is in order.