Be atomic illustration


Be atomic

Complexity needs to sit with the human.

Large Language Models can’t work with monoliths. It’s up to the human to be in charge of the strategy and the machine to supply the atoms.

This idea has been popular within machine learning since the 1970s when Hofstadter coined Hofstadter's Law.

For Large Language Models this relates to the number of ‘tokens’ of content they’re able to “see”. For human conversation this is the amount of information we can retain in our working memories both when listening to somebody and responding to them.

To use a tangible example you wouldn’t expect a dev to be able to tell you how to build an app in one go, but they’d be able to answer a question about a specific component or function and help suggest what the next piece of the puzzle would be to solve.


In the Great Pacific garbage patch there is a public perception that there are floating islands of junk. Instead the litter is low density, this makes it harder for humans to detect and more dangerous for marine life. Can you write me an initial outline for a policy paper I’ll be presenting at an international symposium about how governments, plastic producers and civic groups can collaborate to remove this pollution. The outline should be 7 bullet points. We’ll build out the paper using those points.


Write a policy paper on how plastics within the Great Pacific Garbage Patch cause an existential crisis for marine life.