ML/AI product experiments

[vg16855]

Hard no. This is just bloat. People who want to use AI tools will use their own. Development resources are better spent on features that help people actually think.

[werererer]

Hard yes, only because AI currently has a lot of privacy related negatives, doesn't mean it will always be like that. if modularly implemented, i think it is great, even if it is "only statistics". I mean search engine is just an algorithm right, and still people rely on it. This is just search engine++

[vanyatwo]

Double hard yes - we should be able to connect Anytype database to own LLMs and train our own digital twins.

[alexkkoo93]

Triple hard yes. I'm actually really liking Notion's native AI and the ability to search through all my notes within the app so seamlessly.

Having this option for AnyType and knowing my data is safe would be a massive W

[flypenguin]

Gooood no please. Really. "Experiments" ... that product is by far not polished enough to sink resources in weird "AI experiments". In if you simply do another chatbot in the end – gooooood no.

Good search: Okay. That's a concrete use case, which might be solved using AI.

Starting with AI (like "hmmmm ... what could we do with AI in Anytype?") -- please no. Let anyone who wants "AI" make a suggestion which problem that apparently solves.

[robotflaqd]

Very hard no (at this point). There are better and more important things to focus on developing at this point than an LLM. There is a topic on the community forum that sums it up real nicely (https://community.anytype.io/t/roadmap-for-local-ai-llm-intergration/22881). This is either going to be half baked or it's going to take up all the time that is better spent developing other features that are highly voted on the feature request page in the community (spoiler: LLM is not one of them). Come back to this only after the app comes out of beta.

[alexkkoo93]

Gooood no please. Really. "Experiments" ... that product is by far not polished enough to sink resources in weird "AI experiments". In if you simply do another chatbot in the end – gooooood no.

Good search: Okay. That's a concrete use case, which might be solved using AI.

Starting with AI (like "hmmmm ... what could we do with AI in Anytype?") -- please no. Let anyone who wants "AI" make a suggestion which problem that apparently solves.

I'm confused. Are you for or against AI developments? I'm not sure what "Good no" means vs. "Good search" lol.

My suggestion is for AI to employ RAG the same way Notion does. That is a problem that I feel only notion sufficiently solves

[flypenguin]

I'm confused. Are you for or against AI developments? I'm not sure what "Good no" means vs. "Good search" lol.

oh, true. to be exceptionally clear: F**K AI!!! 😆 . Reasoning: Having AI "because AI!!" is probably nice to hunt for VC money, it does not solve any real-world problem. "Good search" is a problem, which might be solved using AI, and in the end who cares how it was solved – as long as it is solved.

I just (literally a couple of minutes ago) posted a rather lengthy post in the forums on what I really, really would like to see for Anytype soon – that is smooth, nice, intuitive behavior and really good keyboard support. To me, "AI" would be something I would actively "oppose" (to be clear: I have absolutely nothing to say here, I just start liking anytype more and more each day, so I start voicing my opinions here 😉) since in my experience it usually takes a ton of resources and delivery incredibly sub-par results.

Also I prefer to operate on tangible goals, and "having AI" ... well, very tangible ;)

[robotflaqd]

AI shouldn't even be considered before formulas, tabs, linked/synced blocks are added to Anytype.

[silpol]

to me, having 'AI just for sake of AI' is rubber-stamping VC funding applications of last years.

what could be positive examples (these are examples!):

• having constantly-learning translation engine, which is used offline (there are others, but this one will learn with your own focus and traits)
• user's own tools to deal with images and videos, from classification to small editing
• more long-tail actual cases ?

[ebanDev]

AI shouldn't even be considered before formulas, tabs, linked/synced blocks are added to Anytype.

I totally agree, furthermore when features as "simple" (putting quotes because I dont mean to disrespect the devs who are doing an amazing work <3) as ordering tags for example are not available

[vanyatwo]

Self-hosted notion alternative with self-hosted LLM integration could bring a lot of devs into the project.

[Heckinchonkeires]

I think I'm mostly going to echo things already said here and in the forum, but I feel strongly enough about this to want to add my voice. I want to assume in good faith that this is on the roadmap primarily to placate those that have bought into the genAI hype and are asking genAI be included for the sake of it. That being the case: in my mind this issue should be moved to the backlog and renamed to something like "Generative AI experiments", and "AI assistant" should be removed from the list of highly requested features in the docs. If it's important to explore machine learning in general and outside the context of genAI, then that could be a new issue on the roadmap.

I wouldn't say that genAI should never be a part of Anytype under any circumstances. However, it needs to be approached as an answer to the question "How can Anytype solve [problem]?" and not as a question itself: "How can Anytype include generative AI?" "AI assistant" as a highly requested (read: planned) feature in the docs and this issue's position on the roadmap communicate that Any is approaching genAI as a question. More often than not (at least from what I've seen), that approach leads to throwing genAI at imaginary problems, or at problems that could otherwise be solved in ways that are better—and, importantly, less costly—from both a business perspective and an environmental perspective.

[robotflaqd]

I love how this has been pushed to the backlog. There are many other things that Anytype need and that would take up R&D before AI is needed.

Would love to see AI implemented for Anytype, but definitely not now as I'd love to see many other things get developed first. Maybe revisit this once the Anytype passes version 1.0.