I am currently planning to participate in GSoc 2024 and I am particularly interested in implementing AI feature that allows users to summarise their notes directly in Joplin.
I am interested to know few things:
If the summarisation feature is implemented, how many notes do you think you will need to summarise on daily/weekly basis?
Do you need to this feature to summarise multiple notes at once?
Are you going to summarise all of your notes or only some of them?
If possible, can you try to give me a percentage of your notes you guys think that need to be summarised?
I am looking forward to your answers, they will help me a lot! If some of my questions are not clear, feel free to tell me!
Note: You do not have to answer all of the questions
I'm a new Joplin user, but have been using other note-taking software for many years.
I have to be honest, I've never had the desire to summarize a note before.
However, if I were still a student, I can imagine that this function would have been useful for me and I would have used it. As a student nowadays, I would probably take notes during lectures with a tablet and pen. Then I would create a summary of all the lectures of the day in the evening. At the end of a semester, I might have made a summary of all the lectures from a course to prepare for the exam.
But if I would take notes during lectures, I wouldn't do it with Joplin, instead I would use OneNote or Goodnotes. Creating handwritten notes in Joplin is unfortunately not nearly as sophisticated as in other apps.
I really like Joplin partly because it doesn't have anything like AI features included. More like a classic app than jumping on the trendy AI bandwagon, which btw would be a privacy nightmare so I think goes against everything Joplin stands for.
(Let's be real though here for a second and acknowledge that most AI is not AI but the word is now substituted when people just mean complex algorithms, ie handwriting to text does not necessarily mean or need and probably isn't AI - it's been done for decades!).
I think a feature like you're after could definitely be developed as a plugin... but I don't see how it would need be AI nor do I think a private individual would of course have the resources for that anyway!
What I'm trying to say in a long winded way (apologies!) is that I feel like AI goes against the spirit of Joplin and the individual freedom it gives one.
I agree on giving users the individual freedom to choose whether to opt in to AI features, whether developed externally or as part of Joplin.
On privacy, the Jarvis AI assistant plugin supports custom models. Most are offline, free, and open source. More model criteria like those and setup instructions are included in the Jarvis guide.
About the proposal, users can already annotate summaries to notes with the Annotate note with Jarvis command. I wonder if Jarvis or its features can be referenced or extended to support the proposal's expected outcome.
Thank you for answering my post and for your input! I do understand your issues with AI being overhyped and a buzzword.
My intention in implementing AI features is not to follow the trend but to increase productivity and decrease users' mental effort. For example, in some posts here in the community, a user said that he has ADHD and that summarizing and tagging notes may help him with organizing notes.
You could use graph-based algorithms to create an extractive summary. However, I am trying to create an abstract summary in my proposal, so I need to use ML for that. As I am thinking right now, I might also give users an option to create extractive summaries. Also, you don't have to worry about privacy at all. It's all happening on the client side.
I am currently proposing to integrate it with the core application. Still, it might also be a good idea to propose it as a plugin and then decide with the Joplin team which is the best option (taking into account the community's opinions and perspectives ).
Thank you for your input and for providing information about Jarvis! The problem with Jarvis is that it only uses LLM models and works only online by providing API keys, except for semantic search. Instead, you can use lightweight models or non-machine learning algorithms to achieve some simple AI features.
I understand that you can use Jarvis offline, but users must create that themselves and the models have to be OpenAI-compatible API. It is more work for users, and some of them are not easy to set up. For example, GPT4All is marked as difficult.
This made me realize that I might need to compare my solution in my proposal to Jarvis! Thank you for that!
That makes sense, thanks, and doesn't sound so nefarious to me!
Personally I think a plugin is the best route to keep Joplin streamlined and within its original intent. Plugins are great because they allow expansion into all sorts of realms like this.
).