exploration / practice / experiments -> experiences

experiences are knowledge

and knowledge needs to be written down (to be cached)

written down knowledge can be further distilled

and it can serve as instructions when seeing the similar problem again

code is a distilled form of knowledge

code is a skill, a language component

that can be reused

in reinforcement learning, experiences are encoded as value function and policy function

they are numbers, bounded to data structure settings

and thus not easily reusable

if experiences are encoded as prompt or / and code

it can be easily reused in different env

knowledge & code are also core of understanding

people don’t understand a thing until he can written down the knowledge or build tools to explain it

<
Previous Post
meta of meta, skills
>
Next Post
search as a core capability of learning intelligence