How I'm Using AI *WITH* My Obsidian Vault
Vložit
- čas přidán 7. 03. 2023
- To try everything Brilliant has to offer-free-for a full 30 days, visit brilliant.org/FromSergio/. The first 200 of you will get 20% off Brilliant’s annual premium subscription.
Monthly Favorites Newsletter - geni.us/XJikV
// Gear
🎵 Where I Get My Music - www.epidemicsound.com/referra...
🎥 My Camera Gear - kit.co/fromsergio
// Want to support me?
☕ Buy Me A Coffee - ko-fi.com/fromsergio
Some of the above are affiliate links, which means I get a small commission at no extra cost to you, which helps support the channel.
This video was sponsored by Brilliant. - Věda a technologie
To try everything Brilliant has to offer-free-for a full 30 days, visit brilliant.org/FromSergio/. The first 200 of you will get 20% off Brilliant’s annual premium subscription.
Can you link your child content brainstorming template?
It doesn't seem to be included with the plugin.
The "Querying my brain" use case is exactly what I'm looking for. I want to dump a bunch of information in my usual messy, unorganized way, and then be able to ask the AI about anything contained within and have it regurgitated in an easy-to-digest way.
Same here, just not too keen on plugging my entire vault to openAI, have a video coming out on it soon!
NucliaDB might be perfect for this. You’d have to use its webui however, at least until someone writes an Obsidian plugin, they do have an API, hint hint 😁
I agree with you@@FromSergio. How about plugging in the Mistral7B model which can run locally on your computer? And by the way, how is that video you've mentioned coming along?
Hey Sergio!
I think I can finally articulate why your videos are so good. How you present your content feels like you are taking the viewer step-by-step, explaining the overall idea, and then diving into the details but just enough and not too much. This allows a very clear understanding of the content, which also improves as you keep watching the video having the last part sounds almost too good to be true.
That being said, I hope you believe me when I say I was almost sure the following video was going to be about AI; it appears it turned out to be even better than I expected!
Lastly, I will ditch my own implementation of my PKM system. I will recreate it using all the tools and techniques from your obsidian video collection. I was waiting for the AI stuff because this is precisely what I have left to implement in my design. Thank you very much, friend, going to binge-watch all of your videos very soon.
Tomer!! Good to hear from you :) That's so kind, it makes me proud to read it! Super glad you feel that way. As for your PKM system - as long as your system works for you, that's what matters! Have a great weekend friend
@@FromSergio 💪❤
100% agree.
Hi @tomer.nosrati, can you comment on what your insights are on improving your PKM system? I am on a similar journey.
@@d3lirious definitely! DO DAILY NOTES!! You have no idea how much of a game changer it is. Everything else became 2nd priority. And don't waste time on many templates, let them grow out of repetitive manual use until you're manually consistent with your format.
The impact it had on me is priceless and life-changing.
Wow, the template short cut hotkey is amazing. I am impressed with the quality of this video. This video is packed with good stuff that need to watch a few times. 🎉❤
Glad it's helpful! Thanks for the coffee :)
Wow! The idea of querying and generating new ideas based on the contents of your vault really sounds like a dream. Asking chatgpt to show you connections of your notes that you did not realize previously. Can't wait to see your video and hope you get it working soon!
Slowly working on it! Super excited to see where this is going :)
Me too! Im waiting for that
This is fantastic! I really like the idea from 12:05 . I think it would be great to have your vault as a conversational partner for writing/thinking. You could even leverage the journaling color chart from one of your other videos and query your vault asking "What are 3 things that the days colored green have in common?" so you could try to replicate those conditions for the future.
Excellent idea Hugo! I hadn't thought of that and have just added it to my list :)
@@FromSergio Happy to help! :)
obsidian vault for our input and knowledge base?
I think it can be executed by using python llama index in python
dont know are there python plugin for obsidian
I am watching your videos already several months and I love it. Thank you for valuable content and another Obsidian geek as myself.
Thanks Ondrej! It means a lot :)
I was waiting for this video! Thank you!
Hope you enjoyed it! :)
Pushing all your notes into the AI and querying about it looks like a dream!
As far as I know, the limit is about 4000 tokens (3000 words), to be shared between the prompt and the completion. That is, roughly 6 pages of text.
The question is: How do you plan to "train" the AI with your vault that certainly contains many more than 6 pages of notes?
Using Embeddings. The problem is the format of my notes, the model (for now) needs it to be a certain format. I haven't had time to fully explore this but I will in the near future and will make a video on it.
@@FromSergio Embeddings gives any block of text a "signature": a 1536 dimensions vector (latest model: text-embedding-ada-002)
Quite impressive to know this vector alone can be used to find similitudes with other related notes, applying the dot product of two vectors (cosine similarity), with a result ranging from 1 (max similarity) to -1 (quite distant).
There's an Obsidian plugin for that as you probably know: "Smart Connections"
I bet this is also the "technology" behind the napkin one addictive app, which present all your similar notes in a fancy manner, without you having to classify them.
Bonus question: Is it still worth it to manually tag or link our notes, if the AI can do it for us?
Now, according to openAI API, they have another end-point to train AI with your own data set: the Fine-Tuning model. After the training, you don't have to manually put your notes into the prompt as a context, because it's already in the "AI brain" by default.
Quite exciting, but it looks quite daunting to feed the AI with 500+ prompts with their respective expected answers. Also, it doesn't look like this kind of training can make the AI capable of answering interesting inputs, like: "Reminds me what are my favorite insights about learning", or like "Based on my notes, give me a list of interesting facts that intersect between learning and teaching".
As with any tech, the question is: how do we want to use it?
With a PKM like Obsidian, full of 1K+ notes: what kind of questions would be interesting to invoke?
It looks like combining AI with Obsidian could lead ultimately to a closer second brain, being able to resurface any thought we once had that makes sense depending on the context at hand.
@@FromSergio really looking forward to this
Thanks. I was wondering which was the best AI plugin and you have answered that. Take care.
You as well! :)
Yay! A blast of a video. I do like your obsidian-gpt workflow, I will reproduce it. I just read yesterday the email on chatgpt 3.5 that openAI sent to us. I think now is a great moment to integrate chatgpt to my vault. I was already using it to summarize some contents throughout Reader Ghostreader function.
By the way, it is also in my plans retrain a text generator model with my texts. I am quite lucky because I am and have been an active user from certain Internet forum since 2003, thus, I webscraped all my threads and posts on that forum giving birth to one big corpus of my own text/way of writing. I'm thinking about take all the posts after 2010, and longer than 5 number of words to retrain some text generation model. This project has priority 2 nowadays :)
Thanks for the video, man :)
YES! Love the priority 2 reference! That sounds great :)
Thank you. Have been looking for tools to do what you’ve shared. Also wanting to train long term too. Will be curious on what you do.
I'll be sure to share it!
Fantastic video as always. Options in text generator plugin seems to be a bit overwhelming, so I guess there is going to a bit of a learning curve
Options are just what you see on the command pallete, they're mostly just different ways of adding a template
As always a very practical and easy to understand explanation of the powerful tools available with Obsidian. Keep up the great work.
Thank you :) Appreciate it!
That definitely does not sound at all stupid, it sounds cutting edge!!! What you have presented here is game changing. Thank you!
Hi Sergio, I really enjoy your obsidian videos and have been watching them for a while now. Lately, I've become more interested in using open AI with my vault. I find the children notes very helpful for my work. However, I'm having trouble getting the template to work for this. Could you please share the template and also create more videos on how to use my knowledge in the vault? Thank you so much. Markus
Ha! I literally just asked about this on the forum. I’d searched but didn’t see anyone else asking, at least not how I asked. Glad to find this. I really want to VERBALLY ask “Hey Obsidian” without typing or being on the ipad, and ask it questions about my content. Can’t wait for that.
Thank you for the video! This was exactly what I needed, and I have it working as described.
I've been looking into embeddings as well. The problem is converting Markdown notes to text for the embeddings. It may be that Obsidian users are so idiosyncratic that everyone needs his own custom Python program to do the conversion. It also seems to me that embeddings do not really leverage the information contained in Obsidian hashtags. And don't forget that the query itself needs an embedding. However, if a search can be done locally and the relevant content put in a child note, the "Include Children" prompt would be a way to get the content up to OpenAI.
Thanks, Dan! That's exactly what I said to another comment. My problem is also the Markdown format in my notes. I'll have more time to dedicate to this project in the summer, so hope to have a video on it by then! :)
Thank you so much! Helped a lot! 🙈
Thanks! This is really interesting! Can you please share the templates for rewriting the text in different styles?
Thanks! i was looking for a place to get started with this amazing plugin.
AI as a productivity assistant sounds great.
Glad it was helpful! :)
Very complete video!
The only problem I have is that I can't access the links and extract the information.
Happy to see these plugins add support for other chatbots beyond OpenAI.
Excellent and thanks for this - the missing link!
Thanks Christopher :)
I really want to thank you so much for your videos. They are so helpful and I appreciate them a lot.
Thanks for letting me know, means a lot :)
Sorry im new here but iv been trying to find a way to download plugins for like a month now. You are the first person that has done a walk through iv seen. What sight, app, link, ect do I go to download get plugins for my chat gpt 4
Your explanation on how to use Text Generator is exceptional, and it will undoubtedly benefit anyone new to Text Generator.
Thanks Noureddine :)
Great stuff!
Hey Sergio, great video!! quick question - are there any privacy concerns with this? Will openai get access to your notes with these queries? I'm not too familiar with how the API works. Looking to get this extension but want to hear what you think first! Keep up the great work :)
Hi Justin! Yes there is definitely a valid privacy concern here. The API needs to process what you give it. And there's no clear way of knowing for certain what happens to the data that is sent. It's a big subject right now not just for the API but the ChatGPT side of things. I plan on making a video on it. As it stands it seems like privacy is being neglected in favor of training the model at a faster rate. Hopefully this changes soon!
this is interesting, I like the idea of AI doing grunt work. rn I am working on a python toolbox that goes around my vault to make life easier for me hehe
this is beyond exciting, and interacting with your vault like this is one of the best use case scenarios I've come across, wonder if their moderation and safety mechanisms will stand int he way of let's say some research that might have controversial ideas. I noticed that it is capable of just making things up.
I know that chatgpt will indeed stand in the way of more controversial topics which leads me to believe so will GPT3.5
love your explanations. concise and straight to the point!
do you have any tips or ideas for using TemplateR and Text Generator synergistically?
Thanks! Appreciate it! I don't but that's a great thing for me to look into! :)
@@FromSergio awesome! 💪
Hi, great vid. When you say you created the childrentemplate based on the documentation, do you have a link to that documentation? I can't find anything too in-depth.
There's probably a way to store your entire vault in a vector database, like ChromaDB, and that use that in conjunction with the OpenAI API.
This is incredible!
dude train the llm on your notes and make a youtube tutorial or at the very least an article about it! i would love to do something like that, I think this idea interacting with documentation can be revolutionary
I'll do my best :)
Thanks for the explanation, may I ask what theme you are using for Obsidian?
Hey Jas! Minimal theme has been my preference for about a year now!
I prefer Obsidian Copilot which gives you a more traditional ChatGPT interface as a side panel that allows you to read a note while chatting instead of chatting inside a note, and even chat about the note you're reading with a simple toggle.
Different use cases!
Hey @FromSergio, Ollama has become pretty good now to run Llama3 at least the 8B one. Wondering if there was a way to use this and a little bit of RAG on the notes for Obsidian. - I am sorry not sure I am making sense, but on a high level, I wanted to ask questions on all the notes I have within Obsidian without openAI/Google eating away at my notes.
This is so cool. Crazy cool.
Like so many others, I love this video and I also enjoyed you Notion vs Obsidian video lots. Iv'e been trying to get this working, and no luck so far. I already had an account with OpenAI, I installed the Text Generator plugin, created new OpenAI key for it, entered that in the settings for Text Generator, and I keep getting "TypeError: Failed to fetch". I've searched the Obsidian forum and more broadly in general google search and not found anything useful to resolve this. Any suggestions on what I'm missing / how to make this work?
Sergio, great video. I was most interested in your last comment about having a chat bot with your entire vault. Have you figured that out yet?! I'm looking for alternatives also outside of Obsidian. Thanks!
Hi Greg, still working on it on my free time :)
there are alternatives besides obsidian
Great video. The thing i am concerned about when using the plugin is the sharing of my note data. When using it to work with research this might be an issue.
The privacy is a valid concern. There are open sourced alternatives that you can run locally but they're not as competent!
@@FromSergio Could elaborate on those or where to find them?
So if I understand this correctly, you could have a master file that links EVERY note in your vault to then query requests like "brainstorm," and "summarize"
That wouldn't work, at least not yet. There are limits that would preven tthis. Embeddings is what I'm looking at to fully train a model on my vault
Thanks for the video. When I thought about integration of AI with Obsidian my mind went to a chat style feature where it could tell me a summary of my thoughts on something in my vault. Would it be possible given the methods you've outlined in the video to get regular prompts about things I need to think about? For example, if I had a note with line after line of 'thinking prompts', could I use the methods you've outlined above to give me a prompt to think and write about each day?
Interaction between personal vault and AI could lead to questions like: "reminds me 10 insights I had about learning over the last 6 months". Or like "Based on my notes, tell me how learning intersect with teaching".
The "thinking prompt" you're exposing seems to be feasible with pulling a random line out of a dedicated note listing all the good prompts you came across. There's already an Obsidian plugin for that, and there's no AI involved. Or I may have misunderstood what it is you were thinking about.
@@yoricman thanks for your input. Your comment has captured my two wishes: 1) to have AI tell me my thoughts/insights on a subject from across my vault. 2) getting writing prompts from a list in a note. I'm very pleased that there is a plugin for no. 2, do you happen to recall the name? Regardless, thanks for your reply.
What is the reason for adding prompts in the notes? I am new to this, so everything seems overwhelming.
Awesome content! ✨ More please! I’m subbed and sharing with friends 🤓🤓🤓👍
Thanks! :)
11:54 this sounds so powerful
hmmm. When you use “prompts” in the menu, you mean “template” right? For text generator “prompts” are “templates” - the same thing?
everytime i think i love logseq more (i mainly take notes in outline format) the next time i use obsidian i love obsidian more
I know the feeling! But obsidian is perfectly capable of outlining, I do mostly long form with specific note types like notes on courses I take being outline. Of course it's not as outline focused as logseq!
Question: does it cost tokens based on what you put IN to chatgpt, or only what it puts out? This becomes extremely relevant if I’m using the context of very large notes.
It costs dollars based on how many tokens (words) are generated. This includes the input and output.
But it’s only $0.002/1K tokens
Is there a way to use chatgpt to auto link notes between each other? Like reword every iteration of the word AI. Artificial intelligence etc so that it’s a common word across notes and then link them all to each other
Or to find subtopics within the posts and then link then
Sergio, great video. At 9:55, I cannot find the command "Text Generator: Create a Template." I can find "Text Generator: Templates: Generate & Insert." Does this mean a different version? I've downloaded the prompts and am using gpt-3.5-turbo
no yuo wrong
Ma man, great as always
Glad you enjoyed it :)
Does it access the content in those obsidian links that you used for brainstorming?
It does!
In three years we'll be able to talk to an AI in our brain which has all of our notes, and access to all of human knowledge.
I'm already volunteering to be the first test subject it gets implanted into.
Pretty cool to see we can use DallE in Obsidian. Now ai can more easily finish my book on Woodchuck Style KungFu.
Probably sooner than that :) / :(
Great video! I'm new to chatGPT and AI, so I find most of it confusing. For instance, where and how do we even purchase tokens?
You dont, it's deducted automatically on a per-request basis when you trigger the API.
Seems very cool! But when I try to use the Generate & Insert Template/Simplify I get a snake chasing its tail forever. Is this a problem with ChatGPT usage? Is there a way to see some kind of debug info? I've tried re-installing the plugin and restarting Obsidian. (I understand you can't debug everyone's installation, but hoping you can point me to what might help me do the dirty work!)
Thanks for the great videos!
Hi Russell! Hard to tell without seeing the logs but a few questions might help - does this only happen with simplify? Does it work fine otherwise? Do you have the "display errors in the editor" toggled on?
This plugin is a comfortable replacement of copy-paste from the GTP web interface. However it is definitely NOT a way to generate content from a model that has been **trained** on your own graph. Prompting a response of an existing model is very different than training a model on a knowledge base.
Never said otherwise!
Thanks!
Thanks so much for your generosity :)
I'm having issues installing "Default prompts package"
Every time I click on it, it quickly loads and then disappears, the box it is stays, but not text or option to install is visible
Is it possible to use open source LLM with the text generator plugin? If yes, how? Please revert.
I tried Sergio implementing this but could not make it work. I followed every step, but the system does not respond.
What about using llama or alpaca AI in your own desktop pc with Obsidian?
Doable. Last I checked Llama was not as good as 3.5 though especially for my use cases. Have no experience with alpaca.
I look forward to trying that out. I agree - I want my vault to have my own work and not an AI's content but getting ideas like brainstorming and help with title ideas ... heck yeah!
Exactly!
Hi! Did you ever make a video on obsidian commander?
I haven't yet! It's on the list!
Personal AI + Open Gate with GPT Copilot.... Obsidian is becoming the best superapp in the entire world, very quickly.
I am not abel to make it work... Do i have to pay for the subscription on open AI?
Thanks! I tried the "brainstorm from children" that you displayed and it gave me wildly wrong information. Will have to keep playing with it!
Same problem here - So first of all thanks a lot for this very informative video. It works fine for me, except the part with the children notes. I don't get that working. There is a difference in the usage of "CONTENT" and "CONTEXT" between the default templates and the screenshot Sergio is showing at 10:03. Unfortunately also the documentation is not very clear about that point. Could you go a bit more into detail how this works?
i tried ,works fine!
Has anyone had the issue where it says you've hit the maximum request but you have the paid version and you've never used it before? Tried some troubleshooting but haven't had any luck >.>
hey, any of those available models on the plugin is 100% free?
I don't believe so!
I have the latest Text Generator Plug-in installed with the gpt-3.5-turbo selected. However, I can’t find the “Default Prompts Package”. Where can I find this? Thanks, Dan.
You need to install them first, then they will be found under a text generator folder in your obsidian file system! I covered it in the video, is it not working for you?
@@FromSergio Sergio, hi. I'd love to install them, but I don't see them listed under Core Plugins or Community Plugins. Currently, I have an Insider Catalyst License, so I would expect to see all the latest plugins. And I've searched for the plugins. Am I missing something? Thanks, Dan.
I found and installed them!
i have the same vision
Default prompts seems to no longer be available? Either that or my index is outdated.
They're still available on my end
@@FromSergio Maybe it's my region or platform, I'll have to try it when I get back home on my usual windows PC. I'm using a Mac Air in Japan right now so that might be messing with my CDN or something
@@delta-a17 same here. Did you manage to fix it?
@@PedroMaia108 Nope.
Same here, can't find the Default Prompts plugin. Is there any way of sharing the old content?
When will gpt 4 be included
It has just been added, just update the plugin. You need to have access to it though and it's currently on a waitlist basis.
The free tokens expiref after 3 months so now the plugin doesn't work and am trying to learn how to hook into a local machine LLM from hugging face.
Can you generate based on all children, and all the childrens children and so on?
Maybe one day soon but not yet :)
Does anybody knows a free alternative?
here before GPT knew this does exist
intersting for sure, but aren't you concerning about feeding openai all your data?
Absolutely, spoke about it on the last newsletter and am making a video on it soon!
sorry i have an error: You exceeded your current quota, please check your plan and billing det ails.
I miss your obsidian tutorials. Hope you change your mind and go back to Making them. St least the ones you have mentioned you had planned to do
Seems the plugin now has stopped working, couldn't get it to work and even tried regenerating multiple api keys
Chat GPT goes down during periods of high usage very regularly for all free users.
once you go plus you dont actually pay for anything. you still get the same 5 bucks a month free but you have a credit card on file with them in case you go over. you can also limit the cost to 5 bucks and then you get 0 downtime free plan.
but I just limited it to 15 bucks, the limit includes your free 5
@@Pinko-Diamond Not sure how that helps my issue but the Author of the Plugin released a updated version that fixed a bug. I'm also on the paid plan.
Am I the only one that does not manage to make it work? To me it keeps loading and nothing happens...
THANK YOU BUT WE NEED FREE SOLUTION !!
other than the brainstorm feature that seems "vaguely" useful. everything else it utter garbage & point less.
the point of obsidian & other Zettelcastine systems is to help you store YOUR thoughts
i.e. the things YOU have spent YOUR time and effort summarizing & rephrasing in your OWN WORDS. That will never be an easy process (unless we have some neural implant that helps us to instantly proses information & make connections).
...
so by doing wut u r doing, you are completely missing the point of obsidian, it is meant to be a tool that help you store YOUR thoughts & ideas and grapple with them & play with them on the long run. what you are tying to do is entirely eliminating you / yourself from your knowledge.
...
ppl out here trying to make "thinking easy". It's not supposed to be easy. The point of using all these tools is to lessen the minor in-convinces that come with the process of generating "new" ideas & expanding "your" knowledge base. In-convinces like, not being able remember all YOUR thought & understandings, not forgetting the connections YOU have made while thinking about YOUR existing ideas, etc... .
While you are out here storing other's ideas & the connections made by a machine based of those in YOUR "second brain", so no, you are not building yourself/your-knowledge base, you are building a fancy bookshelf filled with others thoughts & ideas.
...
ps, you shouldn't be putting in other's words in your zettel system to begin with, other than "quotes" every single character in your vaults should have been typed by you from your thoughts.
lol mad
Notion AI is available…no plugins, nothing!
And?
@@octain read it and go on to your life mate
@@bernardobotelhomiranda9806 bozo
What about the fact that these AIs make a lot of mistakes?
Inevitable for now. But it's come a *very* long way since davinci 002!
Hey guys, it's your favorite trillionaire Jeff Bezos here, just popping in to remind you all that the class divide isn't so bad! I mean, sure, you might not have a private rocket ship or a gold-plated bidet, but think of all the money you're saving by not having to hire a team of accountants to handle your tax evasion! Plus, who needs basic necessities like healthcare or affordable housing when you have Amazon Prime? Just remember, if you work hard enough, one day you too can become a billionaire and join me in the clouds. See you all at the top!
FOR THE LOVE OF GOD DO NOT UPLOAD YOUR 'SECOND BRAIN' TO A COMPANY. You do realize that when you use the OpenAI API using your notes as input you are giving them all your data right???
this is cool and all but i wish you werent so basic about what you showed. how could this help with fantasy writing or DnD play ..... Mix things up abit from time to time.
I do not do either of those things. I only talk about what I do use.
It's not cheap! Chat GPT has a limit of texts that can be analysed. What about using 3.5 turbo? It doesn't analyse all the text of the value will it? It's just like chatgpt but in obsidian for now.
It is indeed cheap for 99% of applications. I use it daily, roughly 10-30 requests, and my monthly bill with gpt 3.5 will likely be 0.5-1$/month by my Davinci 003 estimates. It does have limitations, it won't analyze a 20,000-word dissertation. But i'm thrilled with it.
@@FromSergio Does it have a higher limit than using ChatGPT though? If you can tell the approx limit, would be great. I love that we can use it in obsidian with a simple setup and looks like it's on a per usage basis rather than monthly plus plan that chatgpt plus offers?