# Full Text: Modern Nostr Index Card-based Knowledge Engineering

> Extracted from `Index_Card_7_5_2023.pdf`

---

## Page 1

•
We have continued to meet and characterize the Index card - Nostr system. 
•
One next step is turning information into Index cards, and determining if it is completely 
compatible with Nostr as-is. Then we will be ready to proceed to the next proximate steps:
•
Composing paths of index cards (extant Nostr messages, extracted from paper, single 
words, etc). An example of this, is a single card as summary/extraction, another note is 
the corpus ID, another note is the word embedding (and then another note to embed the 
tuple, there can be different tuple packages provided). 
•
Using semantic embeddings from LLM (and translations, summaries) to use index cards 
as contextual/semantic bridges. 
•
There are some further possibilities beyond these, however these Nostr-based functionalities 
above (compositionality & LLM-derived embeddings) will provide the setting for discussing 
those. 
•
what edges do we build between index cards when we are parsing a corpus and training the 
LLM?
•
Edge is made by connecting two unique index card ID. 
•
Index Cards
 
  has JSON: Username, Text, etc — then hashes into 32 bytes, SHA256. 
•
Could have ZK-knowledge intermediary.
•
Can connect to BTC (BTC ordinals with Nostr hash sequences). 
•
Nostr giving DOI  
•
what pathways exist between index cards?
•
It will be possible whether or not every index card has total metadata. 
•
how are index cards related to the outcome of the system i.e. Example: Training an LLM 
•
Grouping topics as mycelial network. 
•
The resources are the original papers
•
Nodes of the mycelial network are the vocabulary. 
•
Moving between the papers, not just with a citation edge. Could be with a Syntax side 
(e.g. bridging via a keyword Index Cards) and with a Semantic side (via the 
embeddings).
•
Semantic embeddings as the midpoint between two citations/texts that have different 
syntactic phrasing.
•
Key concept + Summary given context → Fine-graining/Coarse-graining on the 
semantics. 
•
Cards with legible language and translation, as well as semantic embeddings. 
•
Local ontologies and embeddings (e.g. “Surprise” may have some embeddings 
on whole-internet data and a different 
•
“When I use a word… it means just what I choose it to mean – neither more nor 
less.”
•
Information foraging: 
•
https://academic.oup.com/cz/article/61/2/368/1792383
 
  From foraging to 
autonoetic consciousness: The primal self as a consequence of embodied 
prospective foraging

## Page 2

•
https://pubmed.ncbi.nlm.nih.gov/26097107/
 
  Foraging in Semantic Fields: How 
We Search Through Memory
•
https://pubmed.ncbi.nlm.nih.gov/25487706/
 
  Exploration versus exploitation 
in space, mind, and society
•
Relays can hold the information or not. 
•
Can pull notes other relays. 
•
There is a request to delete sent out. No edits, or forced/verified deletion. 
•
Incentives for running relays. Hobby, Identity verification, Business. 
•
Nostr wiki
•
First layer is the people with ability to write primary article. 
•
Second relay enables access, and aggregating it with other information. Can provide 
commentary to primary articles. 
•
Could there be simpler/alternative repository? 
•
Index Cards
 
  enable modular read/write/share of information. Not stored in some large 
blob. 
•
What alternatives exist for compacted specific reference of information. 
•
Nostr in academia
•
Open science
•
Shared knowledge base that is open and collaborative. 
•
Alternative publishing platform
•
Association with Bitcoin/other payment
•
Andrew’s Nostr  flashcards 
•
short: https://github.com/limina1/indextr-principles
•
expanded: https://github.com/limina1/indextr-principles/blob/main/details.md
•
Template working data spec
 
 
•
Index Cards
 
  have an embedding, from any given language model.
•
https://openai.com/pricing
 
 
•
Representation that can be short.
•
Can be used for categorizing information that is later developed. 
•
Agent themselves can read/write/edit/add these cards. 
•
Auto-generated papers — may exacerbate the information overload, diluting the information 
environment. 
•
Don’t need to worry about auto-generated content with Flashcards. Because useless ones are 
simply not used. Useful ones are composed together, and can have patterns (e.g. alternation of 
Observation, Complexity concept, Action possibility). Finding paths. 
•
Zettlekasten and composition/synthesis. 
•
Domain-specific skills (Math, Biomedical, etc)
•
This helps clarify how Complexity is being used — movement and action in this space, 
invoking domains and disciplines as needed. 
•
Which paths are having short/long routes in the Autopoietic PCA?

## Page 3

•
Paths can be analyzed for trivial features (e.g. overall length) or various subtle things (e.g. 
which proportion of links are novel reports).  
•
We want an AI to create a research paper — what is the autopoietic PCA for this? 
•
Review paper — most popular paths. 
•
Novelty search — least traversed paths. 
•
Render the content (path selected) with Figures, can do Audio/Visual


---
*Extraction method: pymupdf*