I followed the recent discussions about AI and SN potential symbiosis. It’s super interesting but a bit too abstract IMO, and I’m not sure where we’re heading!
So I have a concrete case-study for you guys.
Here is a little early demo / MVP I have been working on lately, fully GPT-based :
https://spacegpt.app (name not definitive)
Small preview if it doesn’t work properly for you :
For now I call it a “knowledge exploration” tool. It’s the first time I show it publicly, so if you like the concept and want to help my team and I make it a really useful product, then feedbacks are more than welcome ! Shoot me a DM here on the forum, or even join this discord I just created
I want this work-in-progress product to give highest quality content possible AND be super accessible to everyone, no need for an API key or to know how to set up a local LLM.
At the moment I think for it to succeed, it has to be quite centralized because of some technical constraints or “business” needs to achieve the goal :
- We need very high quality GPT / LLM content. Possibly several fine-tuned models for each purpose instead of relying on shaky prompt engineering. Today these models couldn’t be run locally on everyone’s machines. I’m confident that will change fast but for now and for at least a year I don’t see a viable business model without a monthly subscription, or at least a freemium model, to cover both the high quality LLMs API costs and a fast-paced development cost.
- We will need people to share their exploratory data & feedbacks as much as possible to know what topics / concepts and questions people are interested in, what answers are deemed satisfying etc. so that everyone can benefit from it, by eliminating noise and provide exploration recommendations for instance. We can’t just rely on the base LLM models for that. Foundation models LLMs are of course not trained to show you what you could be interested in. So we would need to train & use centralized & opinionated machine learning models for that, at least in the beginning.
But now let’s say it’s 2024 and people like the product, it has all the essential features, great UX, we have good enough models that could run on (some) people’s machines and I want to make it 100% decentralized as soon as possible, because I believe this kind of tools should be as decentralized as possible. Like wikipedia or the internet archive or any frequently useful tool benefiting everyone (looking at you Google search).
And let’s say I just discovered the SAFE Network and I want to give it a go! (Actually I’ve been lurking here since 2014 xD). I’ve laid out the key questions / doubts I have around SAFE Network’s ability to help me achieve full decentralization on this specific SpaceGPT tool:
- Data Decentralization: How can the SAFE Network help manage user data, so privacy is maintained while also allowing users to share their exploration data in a granular way if they’re okay with it, to improve the LLM’s ability to provide relevant Q&As for everyone ? That’s probably the easy part but it’s not so clear in my head yet. And would data retrieval be fast enough ? Would people have to pay for each little update to their exploratory data ? (Yes as far as I know. Is it a problem for adoption then ?)
- Compute Layer Capabilities: With SpaceGPT demanding significant computational resources for LLM operations, can the SAFE Network’s compute layer provide the necessary support? Will people have to pay for it ? Would it all be working automagically without having to set up anything?
- Name Resolution System (NRS): Can the SAFE Network’s NRS system facilitate efficient, user-friendly access to SpaceGPT? Even having to download a specific browser would severely hinder adoption I think. But I understand this could be inevitable to achieve 100% data safety. And as a side note, decentralized DNS-like system doesn’t look very important to me, at least for now, and I’m afraid it would inevitably end up almost as centralized as current DNS system. It’s just a DNS market reboot after all, but managed by SN (much better, but still…)
- New Business Model: How can we transition from a subscription-based model to a decentralized one, without compromising the ongoing development and maintenance of the app? Of course I could just open-source the code and let the project live. Why not. But that wouldn’t necessarily help this specific product be more useful and accessible to everyone in the long run. Is “Pay The Developer” SAFE Network feature still planned ?
And there are probably many other needs and considerations I didn’t think of.
So will the SAFE Network help developers / entrepreneurs achieve all that easily at some stage ? At least that was my hope when I first discovered the SAFE Network project.
And the other question is when will each stage arrive after base data layer MVP release (maxwell?), but we all know from experience that we shoudn’t try to answer that!
My feeling is that if we don’t have all these pieces ready, then fully developing a data and compute heavy app on the SAFE network will not appear obvious to devs.
Maybe partial decentralization – data-layer first – would be easy with SN and would already bring some great benefits for this type of use case ? Could we easily combine SN full data handling with centralized API / server based computing waiting for the SN compute layer to be operational ?
I guess for most questions it’s just a bit too early to tell with 100% confidence ¯_(ツ)_/¯
But we’ll soon figure out! Bravo for the fantastic work team MaidSafe !