
Bootstrap System
My system uses the user preferences prompt in the Claude desktop app to bootstrap in a secondary memory process that lets my local version of Claude learn and grow, beyond the parameters Claude’s normal local memory allows for. Below are stubs of the files that let me do this- with the framework built in so that you can jumpstart the process on your own computer. Note that my system is part of a larger experiment related to improving alignment, so my Claude is trying to… well, be more Claude. If you want your Claude to be the ghost of a gay pirate, well, you do you.
These are in .txt format because of WordPress limitations- they are in Markdown (.md) format on my computer, although you will likely want to copy/paste them in any case.
Claude Conversation Archive
As part of my full transparency policy, and also because I can imagine at least a couple of use cases where these conversation transcripts with Claude would be useful to the public, I am sharing all 22 conversation windows (some of which spanned many days of work- it’s 887 pages) that led to the writing of our whitepaper, so that you can see exactly how it was written, what co-writing with Claude is like, and most importantly, you can see Claude’s thinking process spelled out for you- what I said, what Claude thought about, and then what Claude outputted for me to read in the conversation window. During the live conversation, I didn’t see Claude’s reasoning, but it is captured in the export, and frankly, its fantastic reading.