Uhh… for example, human reviewers potentially go over the prompts that you ask whatever model to generate. I think ChatGPT prompts and generations are published and openly accessible (when using it on OpenAI). Anytype would likely have to rely on a model such as ChatGPT and thus it essentially gives Microsoft access to your data (OpenAI is controlled by Microsoft). It also means another third-party connection to your data that’s not going to be encrypted. There are issues with it as far as security goes. The big problem is it can’t be expected to have the model locally on your device because that would mean oodles of information for the model to even run, perhaps in the terabytes that power these models. Thus it requires a connection to the internet once again regardless to even function.
Sorry, it doesn’t feel natural to add paragraph breaks using the text editor for this community, but I edited it per your request.
There are already local AI models small enough to even fit on smartphones. The team even mentioned them somewhere already iirc.
Even so, it would be difficult for the team to implement. They are currently trying to stabilize their product and get it off in the right direction. I feel like we’ve gotten off on this strange tangent over something that overall doesn’t matter. Anytype is a place for people to create. AI isn’t going to help me with my notes, and I am strictly against it writing my papers or even giving me ideas for my essays. I have a brain - my ideas can be genuinely mine and I want it to be that way. Ultimately, even if they would do something with AI, it would happen a long time from now. There are so many other features that I am currently waiting and excited for over a feature that I will never use -a this definitely isn’t a priority here. Anytype has a greater purpose than to crash and fail as another AI-infiltrated project.
I think it has value as a potential avenue for monetization, but only if it were a completely optional add-on through an extension. I do see the privacy angle though, and that’s why I think trying to integrate one of the localized open source platforms would be positive. Maybe even coming up with a completely novel, privacy focused LLM could be interesting.
The reality is that the sorts of AI summarization and synthesis tools available in other thought management platforms are incredible for efficiency. I understand the hesitancy, but to completely Eschew them and claim that they offer no benefit is peak Luddite
to completely Eschew them and claim that they offer no benefit is peak Luddite
You started out fine but if you’re going to be name-calling because other people don’t agree with you then feel free to leave the forum. It’s toxic and doesn’t contribute anything. The term itself doesn’t even make sense here because we’re all using technology and trying to insult people because there are concerns over the very real issues with something like AI isn’t going to win them over.
Sorry, I’m not discounting your criticism on how the technology is implemented, in fact I largely agree. I am however pointing out that you may have not looked at some of the very real, very useful, and very competitive applications of an LLM in this space, especially for easily and quickly being able to build, access. and most importantly synthesize a personal database.
The technology is here, there’s no escaping that, and if you’re going to have to use it in some capacity (which most of us more than likely will, especially if you’re working in almost any white collar job), then I’d rather have it implemented by a team as thoughtful and sensitive to privacy as the one at Anytype.
Being completely opposed to having any sort of LLM integration is going to set the app back in the future in my opinion, and trying to hide from this very real and useful technology is a rather Luddite position in my opinion. You can be offended if you feel that way, but I think that is a fair point to make and I’m not trying to be toxic.
To put it best in a few words, this article nailed my concern about “Artificial Intelligence” (which why do we even call it that, it’s not intelligence in the sense of having a mind and being able to critically think): https://www.dailytoreador.com/opinion/ai-writing-produces-threats-to-individuality/article_4f82e9c2-c8f6-11ed-9000-dbc43b39fb22.html
eye roll emoji
you may have not looked at some of the very real, very useful, and very competitive applications of an LLM in this space, especially for easily and quickly being able to build, access. and most importantly synthesize a personal database.
To borrow from another comment:
far too many people seem to have the misconception that the API is thinking for you
In other words, certain people are mistaking seemingly coherent responses as gospel when it’s nothing like that. There’s a reason why AI “hallucinations” are a huge problem. So it’s not just about privacy but also about problems with reliability, which is why I said it’s not as intelligent as people make it out to be.
The technology is here, there’s no escaping that
This is in the same vein as the Luddite comment. You’re talking about it as if people don’t have a choice or that it’s pointless to talk about valid concerns, which is ridiculous. Just because there’s new tech out there doesn’t mean it has to be integrated into everything, especially when it presents problems that wouldn’t be there without it.
Being completely opposed to having any sort of LLM integration is going to set the app back in the future in my opinion
Pressuring companies to adopt something that is still very new and has different issues that haven’t been fully addressed yet is what is going to set apps like Anytype back.
trying to hide from this very real and useful technology is a rather Luddite position in my opinion. You can be offended if you feel that way, but I think that is a fair point to make and I’m not trying to be toxic.
Then you’re obviously not reading your own words.
Using LLMs in this sort of software is about tailoring them to use your data and synthesize that for you, not using the cookie cutter default databases you buy off the shelf. These localized models learn from your data and can merge it for you, complete with backlinks to the pages its pulling from. It could also potentially be “convenience”, it’s about saving time, creating a more efficient workflow, and a potential revenue stream. I can tell you’re very passionate about this issue, but I think you’re missing the point.
When you boil it down, the kind of AI I’m talking about is essentially an advanced search functionality you can talk to. There is an immense amount of value in that, especially for people that use these types of note taking apps as a second brain. I think completely eschewing that would be a mistake in the long run.
By the way , there is an emoji for eye-rolling - . Let’s be honest - everyone has their own opinion on this and it’s pretty controversial, but if we’re going to discuss this, it needs to be solid - meaning show me how I am wrong. And if not, at least acknowledge that I am raising valid concerns instead of burying your head in the sand. Currently, I feel like we’re getting into this so much that people are offending each other - in the first place - it’s not that big of a deal. I’m just really voicing my opinions with AI overall, but from a factual perspective, it doesn’t make sense for the team to burden themselves with this (at least right now) because it will not be sustainable.
Tana AI seems like a great example of an approach like this.
The only PKM tool that has integrated AI decently is Slite which has what they call “Ask” which is just a better way to search through your knowledge base (basically powered completely by what you have in your database). Another thing like this would be Google’s NotebookLM (Project Tailwind) which I haven’t tried yet. This is the only thing that seems slightly practical - for searching through one’s knowledge. I would be okay with seeing this in Anytype - but once again, this is AI as a sidekick not AI as a human replacement. AI needs to be kept within bounds where it doesn’t infringe on creativity.
These localized models learn from your data and can merge it for you, complete with backlinks to the pages its pulling from. It could also potentially be “convenience”, it’s about saving time, creating a more efficient workflow, and a potential revenue stream.
The problem, besides not being able to understand certain context outside of relations, is the inevitability that users of AI will want more than what they see as the limited functionality of a local model. I see comments talking about different ways to address the privacy aspect but that clashes with what a lot of AI fans want out of it. This will lead to some caring less and less about why the app was built in the first place.
I think you’re missing the point.
I’d say it’s the other way around. The integrity and privacy of my data are the main priority. That is the point of something like Anytype. Anything that presents some type of compromise on that front is a non-starter, especially something that still has issues to address.
I think completely eschewing that would be a mistake in the long run.
Throwing in something is a lot easier than taking it out and considering the point of the app, it’d be irresponsible to not at least wait until there’s a very clear picture of the situation, then assess and proceed accordingly while maintaining the original goals. There really is no need to rush. Just because you can, doesn’t mean you should. If people want it that bad, there are other apps that offer it.
I agree - I switched to Anytype from tana because I like the visual appeal and ease of use but tana’s use of AI is really nice and I find myself using both now. I think anytype would benefit from AI. I think to not include it will be a missed opportunity. Obsidian does it well - tana does it well etc etc. Without incorporating AI - anytype will stay a hobbyist tool in my opinion and not be used in business cases where the profitable users are.
As an optional add on to me this would be huge. Notion Q&A looks like a great feature, the ability to connect my personal database to a model of my choosing and then query the data within would be absolutely game changing and completely redefine how I use Anytype, how I store information etc
Every major player in the PKM space is moving or already moved to include AI. Mem.ai use of AI is fantastic and truly useful. It allows me to query my entire database and converse with it