As soon as quantum computers have enough qbits prime factorisation can be done very quickly. Not sure the timeline on that as there are a lot of challenges in the technology and it is hideously expensive, but a lot of the move away from RSA to elliptic curves is driven by readiness for quantum computing.
In agree that this niche is DOA. No offence to OP but the barrier for entry to this stuff is low. I built basically the same thing over a weekend for personal use. React frontend, python server, chroma for embeddings, sqlite cache, switch between open AI and anthropic (I want to add llama for full local execution when I get a better pc). I have a local SPA with named "projects", can configure crawl depth from a start page, I can set my crawl rate, don't have to pay to use it, can choose any provider I want... I'm just one guy and that took a day to get working plus a bit of polish.
I would guess the hardest thing by far in developing the advertised product would be user management, authentication, payments and wrapping the subscription model's business logic around the core loop. And probably scaling, as running embeddings over hundreds of scraped pages adds up quickly when free tier users start hammering you.
My question when deciding to sell something I've built is, if building the service model is harder than building the actual service, where is the value add?
My take on the natural evolution is that collating and caching documents, websites etc for search (with source attribution ideally) is a problem that will I think ultimately be solved by OS vendors. Why sign up for SaaS and expose all your content to untrustworthy 3rd parties, when it's built right in and handled by your "trusty" OS.
In the meantime, I reckon someone more dedicated than me will (or probably already has) open source something like I built but better, probably as a CLI tool, which will eventually reach maturity and be stolen cough I mean adopted by the top end of town.
Ethically I think nothing's changed for centuries in regards to plagiarism and attribution. It gets easier to copy work and thinking, but it also ultimately gets easier to acknowledge sources. Good folk will do the right thing as they always have done.
Regarding efficiency, I think tools like this have a place in making access to relevant and summarised knowledge during general research more efficient, when doing the broad strokes to find areas of interest to zoom in on, when more traditional approaches take over.
Interesting times anyway. I have to give credit to people that try, but I'm taking a back seat in thinking of ideas to productise in this space, as by the time I've thought it through, something new comes along that instantly makes it obsolete.
> the barrier for entry to this stuff is low. I built basically the same thing over a weekend for personal use. React frontend, python server, chroma for embeddings, sqlite cache
Lmfao. God bless HN for keeping this meme going for decades by now.
Your last paragraph really says it all. You haven’t accomplished anything in the space and you’re not willing to try. So you’re just going to hate on everyone who does.
Nobody cares how you would build it because you haven’t. At least not in any form that we can see.
As soon as quantum computers have enough qbits prime factorisation can be done very quickly. Not sure the timeline on that as there are a lot of challenges in the technology and it is hideously expensive, but a lot of the move away from RSA to elliptic curves is driven by readiness for quantum computing.
https://en.wikipedia.org/wiki/Post-quantum_cryptography