#meta #idea Last modified date: 2022-12-05 16:18 Commit: 6 # [[Epistemic status]] #shower-thought #to-digest # Related # TODOs # Obsidian plugins dropping some ideas here also https://forum.obsidian.md/u/louis030195/summary - [x] [[Obsidan OpenAI plugin]] -> https://github.com/louis030195/obsidian-ava - graph brain entropy colourised -> i.e. nodes red have been edited recently, while blue are old ones not updated - [x] automatic tag inference using classification model -> https://github.com/louis030195/obsidian-ava - [x] auto completion using a language model fine tuned -> https://github.com/louis030195/obsidian-ava - [[Epistemocrat]] challenger conversational bot that tell you you are a fucking ignorant idiot while you are writing your thoughts using [[Seeker search augmented conversational bot]] and thus you learn a lot and put your ego away -> https://github.com/langa-me/obsidian-langame - [[Seeker search augmented conversational bot]] that use your google history as a [[Philosophy/Epistemology/Knowledge|knowledge]] engine -> kinda big project, need to crawl and index the pages in real time, and use [[Seeker multimodal CLIP memory]] - [x] bunch of language models hacks edit etc -> with templater https://github.com/louis030195/obsidian-ava - [x] [[Multimodal]] [[Semantic search]] -> -> https://github.com/louis030195/obsidian-search - just text yet, need to impl CLIP etc - Google photo integration (with custom filters, i.e. only share my meal pic for example through a cloud function that run classification/object detection to filter) https://www.npmjs.com/package/googlephotos - use NLP models to fix garbage kindle to readwise broken quotes "But the implica¬ tion that something can be right without being expedient, "or expedient without being rigHt, is the most pernicious error thaFcould"possibly be introHut hTriruFTHaiTextremely reputable philosophers" ~ [[Cicero]] - Model based on my readwise quotes and books, i.e. quote extractor from books - https://github.com/fivol/netwik-obsidian but better - add `publish: false` to all #shower-thought notes - google photo integration https://forum.obsidian.md/t/google-photos-integration/41822 - [[Experimentation]] plugin that let you design run and digest experiments [[Weekly experiment template]] - ouraring.com integration -> maybe just templater hack - timingapp.com integration https://web.timingapp.com/docs/#introduction -> maybe just templater hack - https://github.com/CarperAI/trlx train a model on my vault turn [[Epistemic status]] into a reward function, i.e. model will then generate high epistemic status text :D ```py import trlx def epistemic_status_to_int(text: str) -> int: if "#shower-thought" in text: return 1 elif "#guru" in text: return 1 elif "#shroedinger-uncertain" in text: return 2 elif "#floating-point-error" in text: return 4.99 elif "#socratic" in text or "lindy-proof" in text or "godel-uncertain" in text: return 4 else: return 0 # optimize some reward function model = trlx.train('gpt2', reward_fn=lambda samples: [epistemic_status_to_int(sample) for sample in samples]) # model is a wrapper with some logit preprocessing model.generate(**tokenizer('What is the question to 42?', return_tensors='pt'), do_sample=True) ``` - sync my notion to obsidian maybe just do a some templater hack - connect my twitter tweets to obsidian semantically - maybe suggest to insert link to my tweet on some notes - - Connect all my communication channel to obsidian (discord, WhatsApp, mail, LinkedIn, etc ) keep it private - fine tine chabot to automate myself - https://github.com/Sanqui/discard2 - or playwright hack - [x] Google Search (or any search engine, Google is online, [[GPT3]] is offline - until trained again) -> done, Templater - [x] Semantic search notes similar -> generate a [[Stable Diffusion]] prompt with [[GPT3]] and then generate below . I.E. visual thinking space - let's say you think about trees and it output some nice plant, awesome no? -> AVA - [x] Automatically count how many times I edited a file - Api lexica.art suggest prompt templater or plugin -> done - multiplayer obsidian - i.e. group learning / digesting [[Information|information]] something - [[Multimodal AI assisted knowledge management]] using templater try it - https://forum.obsidian.md/t/graph-colourisation-according-to-brain-entropy/20402 - template that takes a paper url and generate a summary (scrap + GPT3) - template that takes a youtube video and generate a summary (scrap + transcript + [[GPT3]] ) - templater that connect semantically few notes together: say you have a link to a note about "cows" and another about "artificial intelligence" - it generate a text that related cows to AI - a templater or something that makes me do math exercises that are fun by talking to a [[Large language model]] - templater that insert my daily reading time and phsical activity time in my daily notes (timingapp.com) - [ ] edit the add commit template to automatically replace existing values if already exist - turning epistemic status into the category it belongs according to nb of commit (e.g. >5 commit = #schroedinger-uncertain , > 15 #godel-uncertain , > 30 #floating-point-error ) - [ ] obsidian templater: prompt me a first name last name -> generate a profile page of someone given information on [[The Internet]] following a template that have been filled by [[GPT3]] given the information found there - [ ] templater that insert readwise highlight from my wisdom tags/collection into my gratitude daily journey - [ ] Goodreads integration where I can count how many time I consumed x book in audio ebook and physical in obsidian - [ ] https://cloud.ouraring.com/v2/docs#tag/Webhook-Subscription-Routes - [ ] tag specific plugin that uses embedding to automatically tag notes - [ ] ui that show as a timeline my path in obsidian notes because we lose track of context so easily # External links