this repo has moved to https://github.com/NotBrianZach/bza

**TODO gif or youtube example

TO RUN

- TODO gifs or youtube playlist here

- if windows then install windows subsystem for linux

- https://nixos.org/download.html (select appropriate operating system from siderbar)

- git clone https://github.com/NotBrianZach/bzabook2aquiz.git

- cd bzabook2aquiz

- nix-shell (flake is work in progress) todo check if mac and use darwin nixpkgs

- make a pull request implementing pretty much all the features this project currently needs to be a nice user experience

- npm install

- ./book2quiz.sh $OPEN_API_KEY path_2_ur_pdf_here.pdf

- get $OPEN_API_KEY key here if u dont have https://platform.openai.com/account/api-keys

- open an issue detailing why doesnt work

how to read this

numbers are steps in the event loop, letters are steps in a given workflow (which modifies event loop)

a.1 -> step a happens before step 1, but after step 0

1.a -> step 1 happens before step a

1.b -> step 1 happens before step a, step b happens after step a

1-a -> step 1 happens concurrently/asynchronously to step a

Giving gpt3 Short&Long term memory (Event Loop)

1. IF EXISTS, load title&synopsis&rollingSummary from readingList.json ELSE try to grab title, query user C=confirm, string=user inputs title, query user for page range of table of contents if input=([0-9]*-[0-9]*) try to synopsize and validate with user, else ask user input synopsis, rollingSummary=empty string

2. IF EXISTS, load pageNumber, chunkSize from reagingList.json, ELSE pageNumber=commandline param (default 0), chunkSize=commandline param (default 2)

3. feed gpt3 pages[pageNumber:pageNumber+chunkSize], prepending beforeContext&synopsis&title&rollingSummary, appending afterContext, summarize pages[n:n+m]

4. query gpt3 w/synopsis+summary of pages[pageNumber:pageNumber+chunkSize] to generate a new rollingSummary

5. while pageNumber < bookLength, set pageNumber=pageNumber+chunkSize, go back to 3.

6. onExit callback function

Quiz Mode: Start with book pdf, then

1.a. generate quiz,

2.a. display summary of pages[pageNumber:pageNumber+chunkSize] and quiz to the user, record user answer to quiz

2.b. (optional, default quiz&answer) gpt attempts to answer the quiz prints answers, R for user reply to answers

6.a record a log of all the summaries and quizzes

Quiz&Answer Mode: Start with book pdf, then

1.a. generate quiz,

2.a. display summary of pages[pageNumber:pageNumber+chunkSize] and quiz to the user, record user answer to quiz

2.b. gpt attempts to answer the quiz prints answers,

query user-> R for user reply to answers, on other input continue

6.a record a log of all the summaries and quizzes

Query Mode:

1.a query user for question,

1.b gpt3 answer user query,

query user

C=continue to next page,

Q=ask another question, repeat 1.b

r=query gpt3 w/user reply on question answer,

A= append next query input to gpt query at the start of each chunk

B= prepend next query input to gpt query at the start of each chunk

e.g. tell a joke about the following

6.a record a log of all questions&answers

Rewrite Mode:

1.a ask user for character (e.g. socrates) (any string will be accepted)

2.a read pages, rewrite in characters voice

Narrate Submode:

todo use https://github.com/coqui-ai/TTS to generate voice

to narrate gpt response

VoiceDiction Submode:

todo use ?talon? to allow voice input?

Reading List Utility (readList.sh)

store path to pdf and relevant executable to read it

backup&rotate logs

switch between query or quiz mode without losing page context using logs

Design decisions

pdf-extract introduces a bunch of binary dependencies relative to alternative libraries but we want those because they enable ocr on the subset of pdfs that don’t cleanly delineate text

Naming

bza are my initials.

also naive pronounciation sounds kind of like pizza, which is typically sliced into pizza just like we are chunking up books.

and bza is a short three letter word which is not too overloaded and can be invoked easily on the command line.

finally, book starts with B, quiz ends with Z and a is a.

makes total sense.

just bza it!

Inspiration

i had previously kept a reading list with commands like

”””

ebook-viewer ~/media/books/TheDividedSelf2010.epub –open-at 59

xpdf ~/media/books/tcp_ip_networkadministration_3rdedition.pdf 50 -z 200

xpdf ~/media/books/LinuxProgrammingInterface2010.pdf

”””

in a file in my /home/$user/media directory so i could read books from command line and record current position

i had also been looking for technically inclined book club without luck

thought had been bubbling in my head that I wanted to read books alongside gpt3,

i had previously spent several years trying to make multi player choose your own adventure novels a thing (and maybe still plan to?)

i really thought, and think, as a massive wordcel, that computers have a vast potential to create new narrative structures

then i saw this reddit post

https://www.reddit.com/r/singularity/comments/11ho23y/first_post_in_reddit_mistakely_used_a_text_post/

and a within a couple minutes, after some good ole reddit arguing, i started writing this

Pushdown Large Language Models

actually, a final thought, about fundamental models of computation

the taxonomy of computation looks like this

finite state machines -> context free grammars -> turing machines

traditional narratives are particularly simple finite state machines at the level of pages

most choose your own adventure novels are also finite state machines, though they have a bit more structure since they are not purely sequential

the way I wanted to implement multiplayer choose your own adventure novels,

i believe they would have been more akin to a push down automata, or context free grammar,

since the story would maintain a list of invalidated edges (which could also be thought of as a unique class of “intermediate” node that dont branch),

and transitions between nodes could change the choices available to other players

i think there is a similar analogy going on here.

reddit user SignificanceMassive3’s diagram displays a “context free” or “pushdown” large language model

which, much like a regular expression is suitable for parsing text, is suitable for the task of reading along with longer form text