First of all: thanks to the devs for making the best note taking experience I have ever experienced!
At WWDC apple unveiled the foundation models framework. Is this something you guys will implement? As AI gets better Iām hoping we can use it for more advanced research of our notes.
One of the use cases I hope comes is the ability to have AI plow through my notes and give me insight.
Example: tell AI to read my daily notes from the past month and give me insight into my mental health, special thoughts and things that have happened that I may want to dive deeper into.
AI can do this much faster than if I had to do this manually. Twos has done an amazing implementation of this (although itās their own thing, and not Apple Intelligence).
What about privacy? The NotebookLM and NotionAI are all tacking this and may be a good option for your use case. Iād be ok with this if I could completely disable AI features from settings.
Yes, we most likely work on something related to Foundation Models to improve search. E.g. if you search ācatā and you a picture of a cat the note will emerge regardless of the presence of the word ācatā in the text or picture.
We are still studying but Iām pretty sure this is something we canāt do. Apple made available their models running that previously were not accessible to third-party developers. This means that if you are not able to have an LLM session with Siri you will most likely not have it with Bear.
All the Foundation Models available so far run on devices as for Photos image categorization and other models Apple used for years.
That stinks that Apple is not opening up the ability for developers to use AI into their own applications. It would be nice to be able to do a ChatGPTālike query into my notes in Bear.
Isnāt that what theyāre doing with their 26 release? Unless I misunderstood, theyāve released a new kit that letās developers tap into Apple Intelligence. Not sure how it will work in practice though.
Do you mean the LLM prompt sessions? They do have some something but as far as I can tell thereās nothing on the server-side LLM level and is not clear if possible to provide context. As mentioned in the WWDC talk, the LLM model provided on device has only 8 billion variables, and chatGPT comes with 1.8 trillion.
I want to echo a few other users by saying that I am absolutely NOT interested in AI in this app. One of the reasons I have moved away from other apps is because of the widespread and large-scale use of AI as a trendy feature. I just want a note taking app, and Bear is a really great app for that.
I ā¦ could theorhetically be okay with this (not that my approval is needed). As a couple others have suggested, the option to turn it on or off (enable or disable AI app-wide) would be a huge incentive for me to stick with the app, because it would make it clear that Bear acknowledges that AI is not for everyone, and may not be best for every use-case.
I strongly advocate against the implementation of anthropomorphism, conversational AI, and personified personalization for now. The scope and breadth of AI go far beyond simply asking, āhow can we utilize this technology in our app?ā Unfortunately, at this time, there are still so many unknowns about AI which we are learning about as we go.
For the use-case of improving the search function within Bear itself mentioned by @trix180, there are questions I would like answers to: How will it handle encrypted notes (titles, content, attachments, etc.)? Will the processing occur locally or on a distant server (and who would own that server); if it is on a server, where will that server be (environment/economy); who will be responsible for the cooling and electrical costs (the end-users via price increases or something else)? What kinds of encryption or security will the processing itself have? Will the content fed (my content) also be private/secure?
(Iām writing this for the sake of Bear staff to get an idea of user wishes, not to have an argument. But you raise some good points so replying to your comment.)
Iām looking for something more like Notebook LM that queries notes than ChatGPT which will create them. For me, itās about mining information, not creating it.
And I absolutely would expect this to be a paid add on. I donāt think non-AI users should subsidise this at all. In fact, payment would add an extra safeguard so users couldnāt āaccidentallyā turn it on.
Good questions about security and encryption. If Bear ever does this (and itās not a deal breaker for me at all, just a nice-to-have feature), Iād be interested to see how they handle it.
First of all, I have to specify that further experiments with Foundation Models showed that they are not suited for an āAI searchā the way you use other LLMs. The main limitation is that they are running on a device and canāt generate results in a timely fashion. We are still considering them to generate a bag of words or a summary of the notes to aid search.
Locked notes in Bear are E2EE, and we canāt process them even if we want to (and we donāt).
As mentioned other times, we are not going to adopt any AI engine that can compromise usersā privacy. This means we can either run them on the device or we are provided we strong guarantees. So far, we have not found any service satisfying these conditions.
This also counts as a reply for @I.am.alex, I guess.
Thank you for such a clear (and quick!) reply. It makes me glad to know our community is communicative. I apologize if I missed any previous information - I am new here in the community board.
I think there would be tremendous value in having a private implementation of AI summaries, insights etc. For example if I have notes from meetings, chats, etc. under a project hash tag, I would love to be able to get a quick summary of the key points, decisions, insights etc. at the end of the week.
The key would be a private, on device, implementation. Iām not comfortable with taking all those notes and putting them into an online tool that could do this at the moment.
Bear has a unique value proposition for me at the moment, but I suspect there will come a time in the near future where AI enabled notes will offer a significant productivity and satisfaction enhancement.
I would hope the Bear team can implement āAIā in a Bear way, for Bear users, and keep innovating.
Iām really of the diametrically opposed stance. I think this nonsense is actually depriving us of our abilities to reflect and delve deep into ideas and concepts. I strongly believe that people using AI enabled notes will see their introspective abilities and creativity decline, slowly making them into the plagiarism machines that LLMs actually are.
Totally see what youāre saying. I just see it as a tool, a way to do certain things when appropriate.
And of course I appreciate the times when Iām able to reflect and be creative.
Just like there are many tools in the Bear command palette I just wish that there was some way to leverage the new technology that is now available, if there is a truly useful way to do it. Iām not looking for a gimmick. And if thereās not a way to do it, thatās fine, maybe itās too early still.
But it seems like todayās āAIā is able to deal with large amounts of text well, and be of assistance in certain situations.
I suppose Bear has little incentive to be at the forefront of it. It does notes so well, in a lot of ways. But I also want them to be around for a long time, and to me that means steady innovation and creativity, in enabling users satisfaction and effectiveness.
Personally I take notes for variety of reasons, but Iād say most often itās because I need to do something else. Notes are the vital stepping stone, but itās what they enable me to do that is really important to me.
