> Damien Le Moal described his experience with using LLMs. It did find bugs, he said, one that was valid and one that was ""pure and utter crap"". The reason for the latter is that it was dealing with hardware, so the context is not just the code, but also the specifications. The LLM may be logical, but the specifications disagree. He is interested in using Sashiko, but is worried because ""someone is going to have to double-check absolutely everything"".

Translation is the core strength under-girding GenAI/LLM tools, we might include stages in the workflow that constrain the focus to the subsystems where changes are made and create flow-chart type 'visual deltas' (eg with a text-to-image tool like matplotlib or mermaid) to aid mapping out and reviewing all the changes.

If we can map out the state of code as-is, we might suffer Hyrum's Law to create a documenting context of the code-at-point-in-time, too.

K3n.