I saw some folks asking questions regarding the costs of Smart Connections, and getting it set up with a large vault. I have a vault with well over 2 million words currently, and recently kicked on the Smart Connections to see how it would do. The thing took a LONG while to get everything cross-connected - well over 5 hours worth of time. I was fairly concerned it was going to crash out, but nope... it just kept going. By the time it WAS done, and I checked the "usage" area of my costs... the total cost of embedding that whole 2 million + word vault was about 30 cents USD. Yep. 30 cents. So, unless you vault is a lot bigger than mine, getting the embedding done is actually really cheap comparatively.
Really appreciate you sharing that example Ira! Yep, the embedding and connections really costs almost nothing and is so useful to get all of your notes connected automatically.
I have been accumulating notes in my Obsidian vault for several years and am way behind on bracketing the key words, so this video opens up using these two plugins to explore my zettelkasten - just brilliant and not expensive. Thank you so much. This is intelligent use of AI as an extension of human capability, not just churning out recycled ideas with no effort and the like.
Glad to hear that! We have another one on the way showing how I have ChatGPT help go through my emails, update tasks in my project management system, and analyze customer details. I'd love to understand some more of what you want AI to do for you so we can make the most relevant and helpful videos possible!
I love the idea of this as long as I use a separate vault. My issue with this is that you need to make sure you're okay sharing all the information in your vault. If not, this is a non-starter. For example, it looks like OpenAI will use all the text in your notes and that means that each note you open or make will be entered into OpenAI to help continue to train the model. So definitely trying this, but giving it a private vault so it can't access personal files.
I would be concerned about this too, but unlike ChatGPT chats which do get used to train their models, this uses their developer API which is never used to train models and is discarded after it is sent to them. The actual embedding data of your notes is actually stored on your computer once processed, and then every chat conversation you have will not be used by OpenAI for model training. So personally I feel much more at peace with using something like this for personal notes than with ChatGPT directly.
@@heightsplatform you are sure about this? Here's what ChatGPT says about your statement: The statement that data sent through the Developer API is "never used to train models and is discarded after it is sent to them" needs clarification. OpenAI's use of data, including data processed through its API, is governed by its privacy policy and terms of service. While OpenAI takes privacy seriously and offers options for users to control their data, it's important to review the current terms of service and privacy policy for the most accurate and up-to-date information regarding data use. OpenAI has mechanisms in place to respect user privacy and data security, but saying that all API data is discarded or not used for training without qualification may not fully represent these policies.
@@heightsplatform I dont think this is completely true, if nothing has changed, last time i checked openai reserves the rights to look to your data, even with the API, if needed and is available to their servers up to 30 days, making them vulnerable to data breaches as well! Privacy is not a trivial matter and shouldn't be so casually dismissed. As far as i know most of the AI solutions as of yet haven't solved the privacy problem thoroughly.
Loved this video, thank you for sharing your setup. I was wondering if you'd be able to share what plugin you used to get your right sidebar categorized from top to bottom? For example, you have 3 separate plugins running from top to bottom below your calendar. Mine are solely nested at the top in the form of tabs. I much prefer the layout you've achieved. Thanks in advance!
Found you when searching for obsidian content. Would be interested in understanding other ways you organize your vault and how you utilize other plugins!
Thanks for letting me know! I should have my vault more organized than it is currently. Mostly, I dump everything there to be able to use with AI, but I'll keep in mind to make more videos about this in the future!
If you are wanting to explore using AI as a writing partner in your fiction writing then Novel Crafter is hands down the way to go at the moment. Like the obsidian plug ins here it connects with open ai and it uses the context of your novel (your codex of characters, plot outlines, sub plots, chapter text, etc). You can ask it to analyse your plot outline and make suggestions, give ideas to integrate sub plots more tightly, develop characters and a ton of other stuff. To my knowledge it does not use your work to train the AI models. AI is just one part of Novel crafter ,it's also basically a tool like scrivner for writers. You can search youtube for novel crafter and find plenty of videos including the main channel which has a lot of indepth examples :).
@@lisabogle Thank you. I did try Novel Crafter. It was interesting but it didn't quite hit the spot for me. Plus, I didn't want another subscription to add to the list. I've been really enjoying using Obsidian, especially the Canvas function for making a story mood board.
Just install those two plug-ins, paste in all your notes/transcripts and start asking questions about them! I feel that I am only scratching the surface so far. Lots of potential for having it review new work in comparison to what you feel is your best work, helping to plan content schedules, etc.
I've done something similar with my Obsidian Vault (with Text Generator plugin), but I'm concerned about the costs associated with the OpenAI API key, especially where you talk about the Smart Connections plugin. You'd know OpenAI charges as we use, and it counts both the Tokens used as "Input" and as "Output", so if we're queueing our whole library of notes with this plugin, doesn't the cost increase massively? I have a ton of notes in my vault, and I'm concerned that queueing with this with my vault as context will cost an arm and leg if I did it everyday LOL. Please tell me I'm wrong, this is as good as note-taking gets, and I really wish I was wrong. For context, anything above $5 a month is too much for me, I live in third world and forex USD costs too much to not care about spending money on something like this. Thanks!
You are wrong! Costs using this would be very similar to costs using the Text Generator plugin (I tried that as well but found Copilot much more well-designed). The only difference would be the one-time cost to generate the vector embeddings of your notes for the first time. This process uses OpenAI's embeddings model so there are no output token costs, and it only has to be done once for all your notes and then for each new note as you create it. Here is where you can see the pricing which OpenAI currently estimates at 3000 pages per $1 platform.openai.com/docs/guides/embeddings/what-are-embeddings for the text-embedding-ada-002 model, which is what smart connections would use. What then happens when you ask a question about your notes, is that AI is finding the portion of your note that is most relevant to what you said and only inserting that into the prompt input tokens. So this isn't changing the context limits that the AI has so much as helping it optimize retrieving the relevant information for you. For example, a prompt like my example of "what is the most surprising thing I said in my notes" can work because AI assigned meaning to the concepts in the embeddings it created which get stored on your computer, but if you tried to say something like, "summarize everything I have said into 2 pages of content" it wouldn't be able to give you an accurate result as it is only going to retrieve a couple portions of your notes rather than being able to fit the context of all of them. Finally, you don't have to take my word for it. You can set a limit inside your OpenAI account! Go to "Billing > Usage Limits" and you can set a hard and soft limit. If you are worried about it costing more than you expect, set a limit of like $3 or something to test it out and get started.
@@heightsplatform I'm curious about this also, as my vault is 2 million + words, and my dream usage would be for inference of a new note that has a day's worth of Micronotes taken, that it can auto organize them and place them in the correct files so I don't have to dedicate a giant chunk of my time just to that, is this possible? Auto predict category placement?? If at the very least, train it to do so?
@@101RealTalker while some kind of AI agent solution might be able to do that in the future, this isn't something that smart connections or copilot in obsidian would be able to do today.
@@heightsplatform I believe a well-articulated prompt something akin to "skin today's daily no-end infer from your knowledge of the overall v folder titles, which passages would most likely go where, given the author's style of organization/categorization overtime.".... And that would be the beginnings of what is trainable... Both it and you, like many dog owners State, the dog saved them more than anything.
Thank you for this video! I learned a lot. For someone who uses Obsidian for academic purposes, is there a way to find out which parts of the answers it spits out are from which notes? I looked at the info of the answer and it just numbers the notes, but doesn’t show the name of the note it used or anything usable to refer back to the notes, even if I put it in the prompt. I’d like to know where the info is from for academic referencing purposes. Would be great if you know of anything that could help me out!
Glad this helped! Using the new "Create a GPT" feature in ChatGPT Plus this is possible. They allow you to upload a file for the AI to reference and search, and when it gives back results it will cite which portion of the paper/file that is from so you can verify. We actually are the first to release a course about creating custom GPTs that I know of, in case you are interested: www.creatorclimb.com/bundles/ai-elevate-program/preview Note though that I am describing does require a ChatGPT Plus subscription and there has been so much demand since this update came out that they have had to temporarily pause allowing new subscriptions. Unfortunately due to the nature of how vector based search retrieval works with AI, it might be difficult for these plugin developers to have the AI cite a source reliably within Obsidian. Hope that helps!
@@heightsplatform thank you!! This is super useful and I will look into it. Perhaps once subscriptions open up again I can look into creating a GPT based on my literature database somehow, excited to see how academics can use this to their advantage in organising knowledge rather than straight up plagiarising it!
In order to generate the embeddings for note similarity, your notes get sent to OpenAI's API. By default, OpenAI will not keep this information (the embedding data then stays on your computer, not OpenAI's servers), and OpenAI's terms state they will never use data you send via the API for model training. If you were to paste a note into ChatGPT without enabling privacy in your ChatGPT settings first, then this information could be used for future language model training. So interacting with GPT this way is actually more private than using ChatGPT by default. I wouldn't want to paste my private notes to ChatGPT, but I don't mind using the API with them since I know it doesn't stay with OpenAI and isn't used for training that way.
Damn... well there goes the idea that I had. Gotta be fast nowadays :( Anyways, this was the idea that got me truly excited about LLM applications for personal knowledge, the sky is the limit with this area. Really interested to see where it goes.
seems promising, but i find your prompts a bit subjective... Asking for the "most interesting...".... who decides what is interesting? Person A would choose note snippet A over snippet B, being more interesting
It is GPT-3.5 or GPT-4 that decides. The reason you might use a prompt similar to that is to find areas of notes or transcripts that might be good to use in a social media post. It is a basic example and you might come up with some more complex and specific prompts. I wanted to get the idea across how fast it is to uncover things you have written, even if you don’t have the keywords to search for it. You could decide what is interesting to you and specify that criteria in your prompt!
I already use Smart Connections, but thank you for the Copilot tip. The one thing I would avoid, that you use as examples, is a GPT prompt that says 'most interesting' and 'most surprising' and 'most unexpected'... For most content, the AI has no bearing of what is 'surprising' or 'novel'.
I dont get the costs, does smart connections cost u money, everytine it calcultates similar notes in rhe background? Can u still put a limit to costs in open ai?
Great question! The similar notes only have to be calculated in one initial processing, then you can continue to go back and forth between your notes and see similarities. The information gets stored on your computer and you do not have a cost every single time you look at a note since it is already calculated and stored. The page for smart connections mentions the following: For example, if your vault contains 100,000 characters, then the initial processing will cost approximately 50,000 tokens. The current token cost is $0.0004 per 1,000 tokens (as of 2021-08-01) which is estimated to be ~$1 USD for 3,000 pages (assuming 800 tokens per page). This is for the initial processing. As you add new notes, each note will get processed and added with the rest of the notes. For all of my notes the cost was a dollar or two. You can set a limit inside your OpenAI account! Go to "Billing > Usage Limits" and you can set a hard and soft limit. If you are worried about it costing more than you expect, set a limit of like $3 or something to test it out and get started. The cost of the embedding processing is one time for smart connections. Then for chatting with your notes in either copilot or smart connections, it is based on how many tokens used.
OK, thank U for the explanation. 👍 so i guess IT maybe cheaper than the Subscription. And Here WE are with my Last 2 question (sorry): you don't Need an Open Air Subscription Plan to use also chatgpt 4 in obsidian via API? Tokens are Just significant more expansive i guess? And what If i use 2 languages in my vault ( someEnglish / some German notes for example, would that bei a Problem ? Appreciate your suggestions
@@om6418 Maybe I don't use it as much as some would, but it is costing me likely around $1/month. I do also pay for ChatGPT+ so I don't use this as a complete replacement, only for tasks that are related to my own notes/transcripts. Correct, you do not need a ChatGPT+ subscription in order to use GPT-4, however I believe you still need to use the API enough to have paid at least $1 to OpenAI before they will grant you access. I'm not 100% positive if that is still the case (we are a software company with our own AI product "Heights AI", so we've had GPT-4 API access since around the time it was announced). GPT-4 is significantly more expensive. For simple questions GPT-3.5 costs essentially nothing and works well, but for complex tasks GPT-4 is usually worth the extra cost.
🎯 Key Takeaways for quick navigation: 00:00 🤖 Introduction to the AI tool Obsidian and its AI plugins 02:20 🛠️ Installing and setting up Copilot and Smart Connections in Obsidian 05:04 💬 Using Copilot to ask questions about your notes 09:48 🧠 Using Smart Connections to find similarities and connections between notes 11:22 ⚙️ Setting up Copilot and Smart Connections with your OpenAI API key 13:48 🌟 Leveraging AI to enhance content creation and organization Made with HARPA AI
🎯 Key Takeaways for quick navigation: 00:00 *📖 Vorstellung von Obsidian und AI-Werkzeugen* - Einführung in Obsidian als leistungsfähiges AI-Notiz-Werkzeug. - Beschreibung der Einzigartigkeit von Obsidian durch Community-Plugins mit AI-Optionen. - Erklärung der Notwendigkeit eines OpenAI API-Schlüssels für die Nutzung. 02:07 *🔌 Installation und Anwendung der AI-Plugins* - Schritt-für-Schritt-Anleitung zur Installation von Copilot und Smart Connections. - Diskussion über die Funktionalität von Vector-Embeddings für effizienteres Notizmanagement. - Unterschiede und Nutzungszwecke von Copilot und Smart Connections. 05:17 *💡 Einsatz von AI zur Analyse von Podcast-Transkripten* - Verwendung von AI zur Extraktion und Analyse wichtiger Ideen aus Podcast-Transkripten. - Demonstration, wie AI thematische Ähnlichkeiten zwischen Notizen erkennen kann. - Beispiel für die Anwendung von Copilot zur Interaktion mit Notizen auf Basis von AI. 07:38 *🤖 Verwendung von AI zur Verbesserung der Inhaltsqualität* - Einsatz von AI zur Bewertung und Verbesserung der Inhaltsqualität von Transkripten und Notizen. - Möglichkeiten zur Nutzung von AI für kreative Prozesse und zur Effizienzsteigerung. - Überlegungen zum Einsatz von lokalen AI-Modellen für unabhängige Anwendungen. 10:29 *📚 Integration von AI in den kreativen Workflow* - Anleitung zur Einbindung von OpenAI API-Schlüsseln und zur Anpassung der Plugin-Einstellungen. - Vergleich der Funktionalitäten von Copilot und Smart Connections im kreativen Prozess. - Empfehlungen zur effektiven Nutzung von AI für Content-Erstellung und Notizmanagement. 13:09 *🚀 Potenzial von AI für Kreativschaffende und Content-Ersteller* - Diskussion über das Potenzial von AI zur Unterstützung von Kreativschaffenden. - Beispiele für den Einsatz von AI zur Inhaltsanalyse und -verbesserung. - Vorstellung von Heights AI Coach als autonomes Coaching-Tool für Kreative. Made with HARPA AI
It does not. Obsidian stores text as markdown files. If you can copy and paste your PDF text into Obsidian notes, it would be able to see it. If you want to use the PDF directly, I know that ChatGPT Plus can view it.
At 5:04 I show where there is a link and video for the copilot plugin on using it with Llama 2 locally! Issue for most people is this isn't exactly plug and play and requires a very powerful computer.
I keep getting "API key is not provided" every time I wanna use Copilot even tho the setting "Your OpenAI API key" does have a key. Can you go more in detail about how to get started after we install these plugins, step by step?
The problem is that the key is only recognized after a restart of Obsidian. That means save the API key from OpenAI in Copilot and then close Obsidian and start it again. Now Copilot should work.
Yes, and no. I gave more details in another comment. OpenAI gives every account $5 in free credit. You can have embeddings processed for 3000 full pages of notes for about $1 of that credit. Then using gpt-3.5 for copilot or smart connections chat conversations, you'd be able to ask a couple thousand questions before using up that $5 credit. Here is some of what I described in my other comment: For smart connections, this process uses OpenAI's embeddings model so there are no output token costs, and it only has to be done once for all your notes and then for each new note as you create it. Here is where you can see the pricing which OpenAI currently estimates at 3000 pages per $1 platform.openai.com/docs/guides/embeddings/what-are-embeddings for the text-embedding-ada-002 model, which is what smart connections would use. What then happens when you ask a question about your notes, is that AI is finding the portion of your note that is most relevant to what you said and only inserting that into the prompt input tokens. So this isn't changing the context limits that the AI has so much as helping it optimize retrieving the relevant information for you. For example, a prompt like my example of "what is the most surprising thing I said in my notes" can work because AI assigned meaning to the concepts in the embeddings it created which get stored on your computer, but if you tried to say something like, "summarize everything I have said into 2 pages of content" it wouldn't be able to give you an accurate result as it is only going to retrieve a couple portions of your notes rather than being able to fit the context of all of them. Finally, you don't have to take my word for it. You can set a limit inside your OpenAI account! Go to "Billing > Usage Limits" and you can set a hard and soft limit. If you are worried about it costing more than you expect, set a limit of like $3 or something to test it out and get started.
8 месяцев назад+1
I don't know why but smart connections is proving completely useless to me, it can't even handle simple requests like "List all notes in my vault that mention regulatory T cells", it fails miserably and I can't see why!
Smart Connections is referencing the vector embeddings as representations of your notes. In the chat feature the AI cannot list out titles of notes, but it can tell you about content from some of them if they match what you are asking. Instead, smart connections has a panel which you can see I open in the video to show related notes. This panel has a "search" button on it, and when you search a word there, like "regulatory T cells" you will get a list of notes and their titles/links. This is the way I use smart connections the most myself, so that it can act as a better AI search than the built-in search feature.
hmm. Firs time i've come across your vid- been soaking up a shit-fuck-ton of obsidian content the last few months since i downloaded it. Trying different higns out. The one chat-gpt plugin i tried to use was garbage- albei tnot this one . SO i shall give this a shot lol. ty
Really? Been using it for months on multiple computers without issues here. Only tried on Macs though if that makes a difference. Have you reported to the developer?
"Analyze anything I've ever written or said" simply is not possible and frankly false. And why would anyone want that? If you're trying to "sell" me something (even if it's just to watch this video), I'm not buying it.
I explained in the video that you have to place what you have written into the tool, or in the case of something you have said, turn it into a transcript (up to you what you put there). As an example of why someone would want this, for me, it has been very useful to remember ideas I talked about on my podcast, where I wanted to find what I said about it, but didn't remember the words to search. It might take hours to find if I had to relisten to the episodes, but now I can roughly describe the idea and AI can tell me where I talked about it. I made the video because I noticed many people paying for all these different AI products that even combined don't offer what you can do easily here.
Isn't it great how people bait with FREE AI, and then by the time you've watched enough of the video to count as a view, they deliver the subtext "except it's not because it requires a paid for api key." Classic RUclips
Genuninely trying to help you here! As I mentioned in the video, the tool is free, you can use local open source AI models with these plugins, and if your computer is able to run those models locally, the AI is free too without any paid API key! If you use an OpenAI API key like I do in the video because it is much easier, then you pay them only for what you use, which for the average person will likely come out to something like $2/month using it on your notes like this, and would be much more affordable than $20/month for a ChatGPT subscription, where it doesn't have any context about the notes you've written. I understand your feedback, and I really wouldn't want to come across like someone trying to bait you into watching. Obsidian is free, and its rare to be able to simply plugin your API key and use something without paying a monthly or at least a one-time fee for being able to do so. Most software that allows you to use your own API key will still charge you to buy the software first.
But its still not free. To use on your local machine is free, but you need a strong pc or somewhat a technical skill, which takes time. You could name the video like obsidian ai for 2$ to make it honest. But you are a lot better than others because you care about being honest, so thanks for that.
@@Adam.3313 thanks Adam. I should have also mentioned that when you first start using OpenAI's API features, they give your account a $5 credit to use. This will more than cover large amounts of notes. So anyone new to all this and wondering what it will really cost, check your OpenAI account's usage page to see how much free credit you currently have. (They used to give accounts $18, but now they give $5). For anyone new this will be free, if you've been paying for the API for other reasons already, then you'll probably have an idea of potential costs.
@@Adam.3313 and @proteusblack8913 I also updated the description to mention about the free OpenAI credit that all accounts get (plus a link where you can verify what credit you have: platform.openai.com/account/usage or click "Manage Account" from the top right dropdown of your OpenAI account, then click "Usage" on the left of that page if you don't trust the link).
@@heightsplatform You truly deserve the views and subs on your channel. I've extremely rarely seen someone invested so much in honesty and answering. I am really amazed. I wish you tons of success my friend. Keep it up!
Why anybody spends precious screen real-estate on calendar or graph is beyond me, might as well have a water fountain and an old man playing chess on the other side too...like, do you not know what day it is, do you need a picture of your name also so you don't forget that either? lol, jk, but also serious🧐
i would think your privacy is better protected in this case because only the semantic meanings are communicated to the LLM and specific details are not. Of course you still need to be careful with things like passwords etc.
@@larrysteinke1839 I am well protected pal. don't worry. I am a cyber security engineer and i know very well the term "even a dead computer buried underground is not safe". Saw in hand experience how AI/LLM gains leverage through my data. No offence but it won't matter to you perhaps exposing what you are doing everyday by daily logs, how you are planning your day, what sort of triggers you are using in a note, how often you take note, in what context. But its super scary to me. And for password... I am on Arch linux and using built in password store which is signed by my gpg key. And the gpg key is not stored anywhere. biometric lock. So every password may it be plain text it is hashed via my key and no AI nor human can decipher it. :)
@@lynic-0091 to me yes I am the most important person or else I won't do work to feed my belly and stay alive. I don't know about rest of the world and I don't need to know.
Copilot's chat is a bit more flexible as a general ChatGPT alternative, and it also allows you to have a conversation about a specific open note very easily, whereas smart connections is better for chatting across all notes and making the connections. For general chat, copilot also allows you to save prompts that you can use as commands to bring up easily. They definitely do overlap in some ways though, so you might decide that you prefer to use only one of them!
I saw some folks asking questions regarding the costs of Smart Connections, and getting it set up with a large vault. I have a vault with well over 2 million words currently, and recently kicked on the Smart Connections to see how it would do. The thing took a LONG while to get everything cross-connected - well over 5 hours worth of time. I was fairly concerned it was going to crash out, but nope... it just kept going.
By the time it WAS done, and I checked the "usage" area of my costs... the total cost of embedding that whole 2 million + word vault was about 30 cents USD.
Yep. 30 cents.
So, unless you vault is a lot bigger than mine, getting the embedding done is actually really cheap comparatively.
Really appreciate you sharing that example Ira! Yep, the embedding and connections really costs almost nothing and is so useful to get all of your notes connected automatically.
@@heightsplatform hi, which model did you use for embeddings?
@@rolandcucicea6006 I use the GPT3.5-Turbo for cost efficiency
I have been accumulating notes in my Obsidian vault for several years and am way behind on bracketing the key words, so this video opens up using these two plugins to explore my zettelkasten - just brilliant and not expensive. Thank you so much. This is intelligent use of AI as an extension of human capability, not just churning out recycled ideas with no effort and the like.
So glad this helped! Completely agree about the use of AI as an extension of human capability, and not just churning out recycled ideas.
Brilliant video. I'm spinning up on Obsidian. These practical AI examples are helpful.
Glad to hear that! We have another one on the way showing how I have ChatGPT help go through my emails, update tasks in my project management system, and analyze customer details.
I'd love to understand some more of what you want AI to do for you so we can make the most relevant and helpful videos possible!
I love the idea of this as long as I use a separate vault. My issue with this is that you need to make sure you're okay sharing all the information in your vault. If not, this is a non-starter. For example, it looks like OpenAI will use all the text in your notes and that means that each note you open or make will be entered into OpenAI to help continue to train the model. So definitely trying this, but giving it a private vault so it can't access personal files.
I would be concerned about this too, but unlike ChatGPT chats which do get used to train their models, this uses their developer API which is never used to train models and is discarded after it is sent to them. The actual embedding data of your notes is actually stored on your computer once processed, and then every chat conversation you have will not be used by OpenAI for model training.
So personally I feel much more at peace with using something like this for personal notes than with ChatGPT directly.
@@heightsplatform good to know
@@heightsplatform although I don't really trust that a big tech company is telling us the truth.
@@heightsplatform you are sure about this? Here's what ChatGPT says about your statement: The statement that data sent through the Developer API is "never used to train models and is discarded after it is sent to them" needs clarification. OpenAI's use of data, including data processed through its API, is governed by its privacy policy and terms of service. While OpenAI takes privacy seriously and offers options for users to control their data, it's important to review the current terms of service and privacy policy for the most accurate and up-to-date information regarding data use. OpenAI has mechanisms in place to respect user privacy and data security, but saying that all API data is discarded or not used for training without qualification may not fully represent these policies.
@@heightsplatform I dont think this is completely true, if nothing has changed, last time i checked openai reserves the rights to look to your data, even with the API, if needed and is available to their servers up to 30 days, making them vulnerable to data breaches as well!
Privacy is not a trivial matter and shouldn't be so casually dismissed.
As far as i know most of the AI solutions as of yet haven't solved the privacy problem thoroughly.
Loved this video, thank you for sharing your setup. I was wondering if you'd be able to share what plugin you used to get your right sidebar categorized from top to bottom? For example, you have 3 separate plugins running from top to bottom below your calendar. Mine are solely nested at the top in the form of tabs. I much prefer the layout you've achieved.
Thanks in advance!
Haven't looked into it in a while, but I'm using the minimal theme plugin. You can drag a tab lower on the sidebar and create a new section.
Found you when searching for obsidian content. Would be interested in understanding other ways you organize your vault and how you utilize other plugins!
Thanks for letting me know! I should have my vault more organized than it is currently.
Mostly, I dump everything there to be able to use with AI, but I'll keep in mind to make more videos about this in the future!
I know there is a way to use smart second brain in local mode, but it's tricky to install.
I hope you can make a video about it.
This sounds really useful for planning a large multiple storyline novel with a large cast of characters.
Yes, that is a great example use for this!
If you are wanting to explore using AI as a writing partner in your fiction writing then Novel Crafter is hands down the way to go at the moment. Like the obsidian plug ins here it connects with open ai and it uses the context of your novel (your codex of characters, plot outlines, sub plots, chapter text, etc). You can ask it to analyse your plot outline and make suggestions, give ideas to integrate sub plots more tightly, develop characters and a ton of other stuff. To my knowledge it does not use your work to train the AI models. AI is just one part of Novel crafter ,it's also basically a tool like scrivner for writers. You can search youtube for novel crafter and find plenty of videos including the main channel which has a lot of indepth examples :).
@@lisabogle Thank you. I did try Novel Crafter. It was interesting but it didn't quite hit the spot for me. Plus, I didn't want another subscription to add to the list. I've been really enjoying using Obsidian, especially the Canvas function for making a story mood board.
This is by far one of the most helpful videos I saw for this integration, and I really need it for my thesis. Thank you so much for sharing
So glad this helped and good luck on your thesis!
Great video and instructions. Subscribed ✅
Thanks for the sub! Glad it was helpful!
What? This video is game changing. Thank you! 🙏
Oh my I need the Intro level to this app! I see a rabbit hole coming with possibilities!
Just install those two plug-ins, paste in all your notes/transcripts and start asking questions about them!
I feel that I am only scratching the surface so far.
Lots of potential for having it review new work in comparison to what you feel is your best work, helping to plan content schedules, etc.
Hey mate, great ideas for obsidian; in terms of rabbit hole - may I ask what theme you are using? 😉
I'm using the Minimal theme/Minimal theme settings and then Make.md and Calendar plugins which make up most of the look.
Smart connectons is a nice plugin, but I won't use it until it can connect to Local models and/or OpenRouter, as the copilot plugin can.
I've done something similar with my Obsidian Vault (with Text Generator plugin), but I'm concerned about the costs associated with the OpenAI API key, especially where you talk about the Smart Connections plugin. You'd know OpenAI charges as we use, and it counts both the Tokens used as "Input" and as "Output", so if we're queueing our whole library of notes with this plugin, doesn't the cost increase massively? I have a ton of notes in my vault, and I'm concerned that queueing with this with my vault as context will cost an arm and leg if I did it everyday LOL.
Please tell me I'm wrong, this is as good as note-taking gets, and I really wish I was wrong. For context, anything above $5 a month is too much for me, I live in third world and forex USD costs too much to not care about spending money on something like this. Thanks!
You are wrong! Costs using this would be very similar to costs using the Text Generator plugin (I tried that as well but found Copilot much more well-designed). The only difference would be the one-time cost to generate the vector embeddings of your notes for the first time.
This process uses OpenAI's embeddings model so there are no output token costs, and it only has to be done once for all your notes and then for each new note as you create it. Here is where you can see the pricing which OpenAI currently estimates at 3000 pages per $1 platform.openai.com/docs/guides/embeddings/what-are-embeddings for the text-embedding-ada-002 model, which is what smart connections would use.
What then happens when you ask a question about your notes, is that AI is finding the portion of your note that is most relevant to what you said and only inserting that into the prompt input tokens. So this isn't changing the context limits that the AI has so much as helping it optimize retrieving the relevant information for you.
For example, a prompt like my example of "what is the most surprising thing I said in my notes" can work because AI assigned meaning to the concepts in the embeddings it created which get stored on your computer, but if you tried to say something like, "summarize everything I have said into 2 pages of content" it wouldn't be able to give you an accurate result as it is only going to retrieve a couple portions of your notes rather than being able to fit the context of all of them.
Finally, you don't have to take my word for it. You can set a limit inside your OpenAI account! Go to "Billing > Usage Limits" and you can set a hard and soft limit. If you are worried about it costing more than you expect, set a limit of like $3 or something to test it out and get started.
I understand now, thank you so much! You're the best!!
@@heightsplatform I'm curious about this also, as my vault is 2 million + words, and my dream usage would be for inference of a new note that has a day's worth of Micronotes taken, that it can auto organize them and place them in the correct files so I don't have to dedicate a giant chunk of my time just to that, is this possible? Auto predict category placement?? If at the very least, train it to do so?
@@101RealTalker while some kind of AI agent solution might be able to do that in the future, this isn't something that smart connections or copilot in obsidian would be able to do today.
@@heightsplatform I believe a well-articulated prompt something akin to "skin today's daily no-end infer from your knowledge of the overall v folder titles, which passages would most likely go where, given the author's style of organization/categorization overtime.".... And that would be the beginnings of what is trainable... Both it and you, like many dog owners State, the dog saved them more than anything.
Thank you for this video! I learned a lot. For someone who uses Obsidian for academic purposes, is there a way to find out which parts of the answers it spits out are from which notes? I looked at the info of the answer and it just numbers the notes, but doesn’t show the name of the note it used or anything usable to refer back to the notes, even if I put it in the prompt. I’d like to know where the info is from for academic referencing purposes. Would be great if you know of anything that could help me out!
Glad this helped! Using the new "Create a GPT" feature in ChatGPT Plus this is possible. They allow you to upload a file for the AI to reference and search, and when it gives back results it will cite which portion of the paper/file that is from so you can verify.
We actually are the first to release a course about creating custom GPTs that I know of, in case you are interested:
www.creatorclimb.com/bundles/ai-elevate-program/preview
Note though that I am describing does require a ChatGPT Plus subscription and there has been so much demand since this update came out that they have had to temporarily pause allowing new subscriptions.
Unfortunately due to the nature of how vector based search retrieval works with AI, it might be difficult for these plugin developers to have the AI cite a source reliably within Obsidian.
Hope that helps!
@@heightsplatform thank you!! This is super useful and I will look into it. Perhaps once subscriptions open up again I can look into creating a GPT based on my literature database somehow, excited to see how academics can use this to their advantage in organising knowledge rather than straight up plagiarising it!
@@Nvtabc you're welcome! Yes, exciting times! We're happy to answer any questions once you start building GPTs.
What screen recording software do you use? The auto zoom ins and outs are nice
Screen.studio! Yes all the auto zooms make life for me and my video editor much easier!
And all your personal notes, daily notes, agendas data etc go to open ai?
In order to generate the embeddings for note similarity, your notes get sent to OpenAI's API. By default, OpenAI will not keep this information (the embedding data then stays on your computer, not OpenAI's servers), and OpenAI's terms state they will never use data you send via the API for model training.
If you were to paste a note into ChatGPT without enabling privacy in your ChatGPT settings first, then this information could be used for future language model training.
So interacting with GPT this way is actually more private than using ChatGPT by default.
I wouldn't want to paste my private notes to ChatGPT, but I don't mind using the API with them since I know it doesn't stay with OpenAI and isn't used for training that way.
@@heightsplatformwow, didn't know that. Thank you. Sounds good as IT seems to be THW Kind of a Plugin i am searching for.
Damn... well there goes the idea that I had. Gotta be fast nowadays :(
Anyways, this was the idea that got me truly excited about LLM applications for personal knowledge, the sky is the limit with this area. Really interested to see where it goes.
Have you seen the new ‘create a gpt’ feature in ChatGPT? It will make basic applications of this even easier. Definitely exciting times ahead!
seems promising, but i find your prompts a bit subjective... Asking for the "most interesting...".... who decides what is interesting? Person A would choose note snippet A over snippet B, being more interesting
It is GPT-3.5 or GPT-4 that decides. The reason you might use a prompt similar to that is to find areas of notes or transcripts that might be good to use in a social media post.
It is a basic example and you might come up with some more complex and specific prompts. I wanted to get the idea across how fast it is to uncover things you have written, even if you don’t have the keywords to search for it.
You could decide what is interesting to you and specify that criteria in your prompt!
So… create a more objective prompt and report back?
Thanks for making this video. Very useful!
Glad it is helpful!
I think you should have mentioned at the beginning that both plug-ins require an OpenAI API key.
he does at 1:01
Agreed. Actually at 0:52 from my phone. (Sorry to nitpick)
what about the Google genesis pro API?
Doesn’t he though??
I already use Smart Connections, but thank you for the Copilot tip. The one thing I would avoid, that you use as examples, is a GPT prompt that says 'most interesting' and 'most surprising' and 'most unexpected'... For most content, the AI has no bearing of what is 'surprising' or 'novel'.
Hey Brian, how is the landing page doing in heights ? Did you upgrade it since covid ?
I dont get the costs, does smart connections cost u money, everytine it calcultates similar notes in rhe background? Can u still put a limit to costs in open ai?
Great question! The similar notes only have to be calculated in one initial processing, then you can continue to go back and forth between your notes and see similarities. The information gets stored on your computer and you do not have a cost every single time you look at a note since it is already calculated and stored.
The page for smart connections mentions the following:
For example, if your vault contains 100,000 characters, then the initial processing will cost approximately 50,000 tokens. The current token cost is $0.0004 per 1,000 tokens (as of 2021-08-01) which is estimated to be ~$1 USD for 3,000 pages (assuming 800 tokens per page).
This is for the initial processing. As you add new notes, each note will get processed and added with the rest of the notes. For all of my notes the cost was a dollar or two.
You can set a limit inside your OpenAI account! Go to "Billing > Usage Limits" and you can set a hard and soft limit. If you are worried about it costing more than you expect, set a limit of like $3 or something to test it out and get started.
The cost of the embedding processing is one time for smart connections. Then for chatting with your notes in either copilot or smart connections, it is based on how many tokens used.
OK, thank U for the explanation. 👍 so i guess IT maybe cheaper than the Subscription. And Here WE are with my Last 2 question (sorry): you don't Need an Open Air Subscription Plan to use also chatgpt 4 in obsidian via API? Tokens are Just significant more expansive i guess? And what If i use 2 languages in my vault ( someEnglish / some German notes for example, would that bei a Problem ? Appreciate your suggestions
@@om6418 Maybe I don't use it as much as some would, but it is costing me likely around $1/month. I do also pay for ChatGPT+ so I don't use this as a complete replacement, only for tasks that are related to my own notes/transcripts.
Correct, you do not need a ChatGPT+ subscription in order to use GPT-4, however I believe you still need to use the API enough to have paid at least $1 to OpenAI before they will grant you access. I'm not 100% positive if that is still the case (we are a software company with our own AI product "Heights AI", so we've had GPT-4 API access since around the time it was announced). GPT-4 is significantly more expensive. For simple questions GPT-3.5 costs essentially nothing and works well, but for complex tasks GPT-4 is usually worth the extra cost.
@@heightsplatform Cool Insights, i will try that out, looking forward to new videos
Do they work with languages other than English as well?
Yes!
🎯 Key Takeaways for quick navigation:
00:00 🤖 Introduction to the AI tool Obsidian and its AI plugins
02:20 🛠️ Installing and setting up Copilot and Smart Connections in Obsidian
05:04 💬 Using Copilot to ask questions about your notes
09:48 🧠 Using Smart Connections to find similarities and connections between notes
11:22 ⚙️ Setting up Copilot and Smart Connections with your OpenAI API key
13:48 🌟 Leveraging AI to enhance content creation and organization
Made with HARPA AI
🎯 Key Takeaways for quick navigation:
00:00 *📖 Vorstellung von Obsidian und AI-Werkzeugen*
- Einführung in Obsidian als leistungsfähiges AI-Notiz-Werkzeug.
- Beschreibung der Einzigartigkeit von Obsidian durch Community-Plugins mit AI-Optionen.
- Erklärung der Notwendigkeit eines OpenAI API-Schlüssels für die Nutzung.
02:07 *🔌 Installation und Anwendung der AI-Plugins*
- Schritt-für-Schritt-Anleitung zur Installation von Copilot und Smart Connections.
- Diskussion über die Funktionalität von Vector-Embeddings für effizienteres Notizmanagement.
- Unterschiede und Nutzungszwecke von Copilot und Smart Connections.
05:17 *💡 Einsatz von AI zur Analyse von Podcast-Transkripten*
- Verwendung von AI zur Extraktion und Analyse wichtiger Ideen aus Podcast-Transkripten.
- Demonstration, wie AI thematische Ähnlichkeiten zwischen Notizen erkennen kann.
- Beispiel für die Anwendung von Copilot zur Interaktion mit Notizen auf Basis von AI.
07:38 *🤖 Verwendung von AI zur Verbesserung der Inhaltsqualität*
- Einsatz von AI zur Bewertung und Verbesserung der Inhaltsqualität von Transkripten und Notizen.
- Möglichkeiten zur Nutzung von AI für kreative Prozesse und zur Effizienzsteigerung.
- Überlegungen zum Einsatz von lokalen AI-Modellen für unabhängige Anwendungen.
10:29 *📚 Integration von AI in den kreativen Workflow*
- Anleitung zur Einbindung von OpenAI API-Schlüsseln und zur Anpassung der Plugin-Einstellungen.
- Vergleich der Funktionalitäten von Copilot und Smart Connections im kreativen Prozess.
- Empfehlungen zur effektiven Nutzung von AI für Content-Erstellung und Notizmanagement.
13:09 *🚀 Potenzial von AI für Kreativschaffende und Content-Ersteller*
- Diskussion über das Potenzial von AI zur Unterstützung von Kreativschaffenden.
- Beispiele für den Einsatz von AI zur Inhaltsanalyse und -verbesserung.
- Vorstellung von Heights AI Coach als autonomes Coaching-Tool für Kreative.
Made with HARPA AI
Does it read content of pdf files?
It does not. Obsidian stores text as markdown files. If you can copy and paste your PDF text into Obsidian notes, it would be able to see it. If you want to use the PDF directly, I know that ChatGPT Plus can view it.
@@heightsplatform Thanks for clearing that up!
Hm, would be much more comfortable with something like this if it was doing the inferencing locally.
Nothing would make me happier, but how?🙏....Every offline thing I research is just so darn complicated :(
At 5:04 I show where there is a link and video for the copilot plugin on using it with Llama 2 locally! Issue for most people is this isn't exactly plug and play and requires a very powerful computer.
Does anyone know if we can use non Open AI models, like Claude 3.5 Sonet (via API)? Thanks!
Yes! Copilot now supports Claude 3.5 Sonnet and other models! You can select them from the plugin settings screen.
What theme are you using
I'm using the minimal theme with the "minimal theme settings" plugin for customization.
I keep getting "API key is not provided" every time I wanna use Copilot even tho the setting "Your OpenAI API key" does have a key. Can you go more in detail about how to get started after we install these plugins, step by step?
The problem is that the key is only recognized after a restart of Obsidian. That means save the API key from OpenAI in Copilot and then close Obsidian and start it again. Now Copilot should work.
As @Emc-it4lg mentioned, restarting Obsidian should fix this for you!
You know NotebookLM?
But is it free though? It seems to require paying for OpenAI even for the basic ChatGPT 3.5, doesn't it?
Yes, and no. I gave more details in another comment. OpenAI gives every account $5 in free credit. You can have embeddings processed for 3000 full pages of notes for about $1 of that credit. Then using gpt-3.5 for copilot or smart connections chat conversations, you'd be able to ask a couple thousand questions before using up that $5 credit. Here is some of what I described in my other comment:
For smart connections, this process uses OpenAI's embeddings model so there are no output token costs, and it only has to be done once for all your notes and then for each new note as you create it. Here is where you can see the pricing which OpenAI currently estimates at 3000 pages per $1 platform.openai.com/docs/guides/embeddings/what-are-embeddings for the text-embedding-ada-002 model, which is what smart connections would use.
What then happens when you ask a question about your notes, is that AI is finding the portion of your note that is most relevant to what you said and only inserting that into the prompt input tokens. So this isn't changing the context limits that the AI has so much as helping it optimize retrieving the relevant information for you.
For example, a prompt like my example of "what is the most surprising thing I said in my notes" can work because AI assigned meaning to the concepts in the embeddings it created which get stored on your computer, but if you tried to say something like, "summarize everything I have said into 2 pages of content" it wouldn't be able to give you an accurate result as it is only going to retrieve a couple portions of your notes rather than being able to fit the context of all of them.
Finally, you don't have to take my word for it. You can set a limit inside your OpenAI account! Go to "Billing > Usage Limits" and you can set a hard and soft limit. If you are worried about it costing more than you expect, set a limit of like $3 or something to test it out and get started.
I don't know why but smart connections is proving completely useless to me, it can't even handle simple requests like "List all notes in my vault that mention regulatory T cells", it fails miserably and I can't see why!
Smart Connections is referencing the vector embeddings as representations of your notes. In the chat feature the AI cannot list out titles of notes, but it can tell you about content from some of them if they match what you are asking.
Instead, smart connections has a panel which you can see I open in the video to show related notes. This panel has a "search" button on it, and when you search a word there, like "regulatory T cells" you will get a list of notes and their titles/links. This is the way I use smart connections the most myself, so that it can act as a better AI search than the built-in search feature.
hmm. Firs time i've come across your vid- been soaking up a shit-fuck-ton of obsidian content the last few months since i downloaded it. Trying different higns out. The one chat-gpt plugin i tried to use was garbage- albei tnot this one . SO i shall give this a shot lol. ty
Some of the plugins out there aren't very good. Out of everything I tried, these two are very capable!
smart connections killed my vault
Really? Been using it for months on multiple computers without issues here. Only tried on Macs though if that makes a difference. Have you reported to the developer?
Yup @@heightsplatform i have 12000 notes in my vault though ? moved my whole evernote archive in there .
What do you mean exactly? My vault consists of 2 million words across 1,000 miles, did you recover?
"Analyze anything I've ever written or said" simply is not possible and frankly false. And why would anyone want that? If you're trying to "sell" me something (even if it's just to watch this video), I'm not buying it.
I explained in the video that you have to place what you have written into the tool, or in the case of something you have said, turn it into a transcript (up to you what you put there). As an example of why someone would want this, for me, it has been very useful to remember ideas I talked about on my podcast, where I wanted to find what I said about it, but didn't remember the words to search. It might take hours to find if I had to relisten to the episodes, but now I can roughly describe the idea and AI can tell me where I talked about it.
I made the video because I noticed many people paying for all these different AI products that even combined don't offer what you can do easily here.
I agree with @heightsplatform
Isn't it great how people bait with FREE AI, and then by the time you've watched enough of the video to count as a view, they deliver the subtext "except it's not because it requires a paid for api key." Classic RUclips
Genuninely trying to help you here! As I mentioned in the video, the tool is free, you can use local open source AI models with these plugins, and if your computer is able to run those models locally, the AI is free too without any paid API key!
If you use an OpenAI API key like I do in the video because it is much easier, then you pay them only for what you use, which for the average person will likely come out to something like $2/month using it on your notes like this, and would be much more affordable than $20/month for a ChatGPT subscription, where it doesn't have any context about the notes you've written.
I understand your feedback, and I really wouldn't want to come across like someone trying to bait you into watching. Obsidian is free, and its rare to be able to simply plugin your API key and use something without paying a monthly or at least a one-time fee for being able to do so. Most software that allows you to use your own API key will still charge you to buy the software first.
But its still not free. To use on your local machine is free, but you need a strong pc or somewhat a technical skill, which takes time. You could name the video like obsidian ai for 2$ to make it honest. But you are a lot better than others because you care about being honest, so thanks for that.
@@Adam.3313 thanks Adam. I should have also mentioned that when you first start using OpenAI's API features, they give your account a $5 credit to use. This will more than cover large amounts of notes.
So anyone new to all this and wondering what it will really cost, check your OpenAI account's usage page to see how much free credit you currently have. (They used to give accounts $18, but now they give $5).
For anyone new this will be free, if you've been paying for the API for other reasons already, then you'll probably have an idea of potential costs.
@@Adam.3313 and @proteusblack8913 I also updated the description to mention about the free OpenAI credit that all accounts get (plus a link where you can verify what credit you have: platform.openai.com/account/usage or click "Manage Account" from the top right dropdown of your OpenAI account, then click "Usage" on the left of that page if you don't trust the link).
@@heightsplatform You truly deserve the views and subs on your channel. I've extremely rarely seen someone invested so much in honesty and answering. I am really amazed. I wish you tons of success my friend. Keep it up!
Why anybody spends precious screen real-estate on calendar or graph is beyond me, might as well have a water fountain and an old man playing chess on the other side too...like, do you not know what day it is, do you need a picture of your name also so you don't forget that either? lol, jk, but also serious🧐
The reason I like that calendar is it can show your writing streak under the days which is nice to visualize. Otherwise, I agree haha.
@@heightsplatform you can say nofap, it's ok, lol
4:40
No thanks, i dont need dumb ai lucrative offer to expose my privacy and personal thoughts.
i would think your privacy is better protected in this case because only the semantic meanings are communicated to the LLM and specific details are not. Of course you still need to be careful with things like passwords etc.
@@larrysteinke1839 I am well protected pal. don't worry. I am a cyber security engineer and i know very well the term "even a dead computer buried underground is not safe". Saw in hand experience how AI/LLM gains leverage through my data. No offence but it won't matter to you perhaps exposing what you are doing everyday by daily logs, how you are planning your day, what sort of triggers you are using in a note, how often you take note, in what context. But its super scary to me. And for password... I am on Arch linux and using built in password store which is signed by my gpg key. And the gpg key is not stored anywhere. biometric lock. So every password may it be plain text it is hashed via my key and no AI nor human can decipher it. :)
Spoiler alert, you're not as important as you think.
@@lynic-0091 to me yes I am the most important person or else I won't do work to feed my belly and stay alive. I don't know about rest of the world and I don't need to know.
@@lynic-0091Security is a real concern and has nothing to do with ego.
It's not COMPLETELY free. Is this video Clickbait ??
I don't get the difference. It seems like smart connections does everything that copilot does and more (can search existing as well as all notes).
Copilot's chat is a bit more flexible as a general ChatGPT alternative, and it also allows you to have a conversation about a specific open note very easily, whereas smart connections is better for chatting across all notes and making the connections.
For general chat, copilot also allows you to save prompts that you can use as commands to bring up easily.
They definitely do overlap in some ways though, so you might decide that you prefer to use only one of them!