Discussion about this post

User's avatar
Gianfranco Gabrieli's avatar

Hi Roger, great piece!

Triggered a thought in my head in these GenAI / LLM days, in which we often assign jobs to AI.

Do the same principles apply to how you prompt your Claude or Gemini?

Following your 3 principles' structure:

1. if it is not concrete (i.e. you are not precise in what you ask and in which format) you can get almost anything;

2. if it is not doable IMO can mean the data set is not in the training data or any other source the LLM can consult, that's when you can see the machine make wrong inferences (hallucinations);

3. if it is not interesting it will make it so. the writing and reasoning capabilities of AI in these days are such that, even without asking it will start to spot holes in your reasoning and propose a more rounded pov on your prompt.

happy to hear your view, great piece again!

Gianfranco

No posts

Ready for more?