On writing less and trying to say more

Core ideas I want to put in this:

• Learning = compression and the ML science around this is well-founded and concrete
• Compression in ML and AI workflows, i.e., asking LLMs to summarise is a powerful thing and context-dependent compressions can be even more dramatic (give an example from one of my maths posts)
• The Dirac quote about poetry might be a relevant addition
• Pictures say a lot of words
• Why compression can go wrong -> what are the dangers of generalisation?
• Generalisation about human issues are often very dangerous ethically