Retrieval-Augmented Generation (RAG) is probably the tech you’d want. It basically involves a knowledge library being built from the documents you upload, which is then indexed when you ask questions.
NotebookLM by Google is an off the shelf tool that is specialized in this, but you can upload documents to ChatGPT, Copilot, Claude, etc., and get the same benefit.
If you self hosted, Open WebUI with Ollama supports this, but far from the only one.
Wow, there isn’t a single solution in here with the obvious answer?
You’ll need a domain name. It doesn’t need to be paid - you can use DuckDNS. Note that whoever hosts your DNS needs to support dynamic DNS. I use Cloudflare for this for free (not their other services) even though I bought my domains from Namecheap.
Then, you can either set up Let’s Encrypt on device and have it generate certs in a location Jellyfin knows about (not sure what this entails exactly, as I don’t use this approach) or you can do what I do:
On your router, forward port 443 to the outbound secure port from your PI (which for simplicity’s sake should also be port 443). You likely also need to forward port 80 in order to verify Let’s Encrypt.
If you want to use Jellyfin while on your network and your router doesn’t support NAT loopback requests, then you can use the server’s IP address and expose Jellyfin’s HTTP ports (e.g., 8080) - just make sure to not forward those ports from the router. You’ll have local unencrypted transfers if you do this, though.
Make sure you have secure passwords in Jellyfin. Note that you are vulnerable to a Jellyfin or Traefik vulnerability if one is found, so make sure to keep your software updated.
If you use Docker, I can share some config info with you on how to set this all up with Traefik, Jellyfin, and a dynamic dns services all up with docker-compose services.
Why should we know this?
Not watching that video for a number of reasons, namely that ten seconds in they hadn’t said anything of substance, their first claim was incorrect (Amazon does not prohibit use of gen ai in books, nor do they require its use be disclosed to the public, no matter how much you might wish it did), and there was nothing in the description of substance, which in instances like this generally means the video will largely be devoid of substance.
What books is the Math Sorcerer selling? Are they the ones on Amazon linked from their page? Are they selling all of those or just promoting most of them?
Why do we think they were generated with AI?
When you say “generated with AI,” what do you mean?
And what’s the result? Are the books misleading in some way? That’s the most legitimate actual concern I can think of (I’m sure the people screaming that AI isn’t fair use would disagree, but if that’s the concern, settle it in court).
Look up “LLM quantization.” The idea is that each parameter is a number; by default they use 16 bits of precision, but if you scale them into smaller sizes, you use less space and have less precision, but you still have the same parameters. There’s not much quality loss going from 16 bits to 8, but it gets more noticeable as you get lower and lower. (That said, there’s are ternary bit models being trained from scratch that use 1.58 bits per parameter and are allegedly just as good as fp16 models of the same parameter count.)
If you’re using a 4-bit quantization, then you need about half that number in VRAM. Q4_K_M is better than Q4, but also a bit larger. Ollama generally defaults to Q4_K_M. If you can handle a higher quantization, Q6_K is generally best. If you can’t quite fit it, Q5_K_M is generally better than any other option, followed by Q5_K_S.
For example, Llama3.3 70B, which has 70.6 billion parameters, has the following sizes for some of its quantizations:
This is why I run a lot of Q4_K_M 70B models on two 3090s.
Generally speaking, there’s not a perceptible quality drop going to Q6_K from 8 bit quantization (though I have heard this is less true with MoE models). Below Q6, there’s a bit of a drop between it and 5 and then 4, but the model’s still decent. Below 4-bit quantizations you can generally get better results from a smaller parameter model at a higher quantization.
TheBloke on Huggingface has a lot of GGUF quantization repos, and most, if not all of them, have a blurb about the different quantization types and which are recommended. When Ollama.com doesn’t have a model I want, I’m generally able to find one there.
You said, and I quote "Find a better way." I don't agree with your premise - this is the better way - but I gave you a straightforward, reasonable way to achieve something important to you... and now you're saying that "This is a discussion of principle."
You've just proven that it doesn't take a moderator to turn a conversation into a bad joke - you can do it on your own.
It’s a discussion of principle.
This is a foreign concept?
It appears to be a foreign concept for you.
I don’t believe that it’s a fundamentally bad thing to converse in moderated spaces; you do. You say "giving somebody the power to arbitrarily censor and modify our conversation is a fundamentally bad thing" like it’s a fact, indicating you believe this, but you’ve been given the tools to avoid giving others the power to moderate your conversation and you have chosen not to use them. This means that you are saying “I have chosen to do a thing that I believe is fundamentally bad.” Why would anyone trust such a person?
For that matter, is this even a discussion? People clearly don’t agree with you and you haven’t explained your reasoning. If a moderator’s actions are logged and visible to users, and users have the choice of engaging under the purview of a moderator or moving elsewhere, what's the problem?
It is deeply bad that...
Why?
Yes, I know, trolls, etc..
In other words, "let me ignore valid arguments for why moderation is needed."
But such action turns any conversation into a bad joke.
It doesn't.
And anybody who trusts a moderator is a fool.
In places where moderator's actions are unlogged and they're not accountable to the community, sure - and that's true on mainstream social media. Here, moderators are performing a service for the benefit of the community.
Have you never heard the phrase "Trust, but verify?"
Find a better way.
This is the better way.
Yes, I know, trolls etc. But such action turns any conversation into a bad joke. And anybody who trusts a moderator is a fool.
Not just trolls - there’s much worse content out there, some of which can get you sent to jail in most (all?) jurisdictions.
And even ignoring that, many users like their communities to remain focused on a given topic. Moderation allows this to happen without requiring a vetting process prior to posting. Maybe you don’t want that, but most users do.
Find a better way.
Here’s an option: you can code a fork or client that automatically parses the modlog, finds comments and posts that have been removed, and makes them visible in your feed. You could even implement the ability to reply by hosting replies on a different instance or community.
For you and anyone who uses your fork, it’ll be as though they were never removed.
Do you have issues with the above approach?
As a user, you can:
If you host your own instance and communities within that instance, then at that point, you have full control, right? Other instances can de-federate from yours.
I recommend a used 3090, as that has 24 GB of VRAM and generally can be found for $800ish or less (at least when I last checked, in February). It’s much cheaper than a 4090 and while admittedly more expensive than the inexpensive 24GB Nvidia Tesla card (the P40?) it also has much better performance and CUDA support.
I have dual 3090s so my performance won’t translate directly to what a single GPU would get, but it’s pretty easy to find stats on 3090 performance.
I’d recommend Colemak Mod-DH, personally - it seems ergonomically superior and switching later is a bit of a pain.