The blog post if you prefer to read:
this post was submitted on 03 Nov 2023
171 points (91.7% liked)
Technology
59377 readers
2961 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
Thank you :)
Im curious to see what sorts of recommended minimum specs there will be for these features. It is my understanding that these sorts of models require a non negligible amount of horsepower to run in a timely manner.
At the moment I am running Nextcloud on some raspberry pis and, my gut tells me I might need a bit more oomph than that to handle this sort of real time AI prompting >_>;
The blog post states:
We build the AI Assistant using a flexible, solution-independent approach which gives you a choice between multiple large language models (LLM) and services. It can be fully hosted within your instance, processing all requests in-house, or powered by an external service.
So it sounds like you pick what works for you. I'd guess on a raspberry pi, on board processing would be both slow and poor quality, but I'll probably give it a go anyway.
Yeah sorry I was specifically referring to the on prem LLM if that wasnt clear, and how much juice running that thing takes.
Some of the other Nextcloud stuff (like that chat stuff) isn't suitable on Raspberry Pi, I expect this will be the same. It's released though, right? Might have to have a play.
You'd be surprised at how little computing power it can take, depending on the LLM.
the AI that nextcloud is offering uses openAI, sign up get a api key and add it. Your ai requests goto the cloud. (and i couldnt get it to work, constant " too many request" or a straight "failed")
The other option is the addon " local llm", you download a cutdown llm like llama2 or falcon and it runs locally. I did get thoes all installed, but it didnt work for general prompts.
Nextcloud will probably fix things over time, and the developer who made the local llm plugin will to, but right now this isnt very useful to selfhosters.
Llamaβs getting pretty damn good, check out phind.com if you havenβt yetβ¦its programming better than GPT-4 supposedly!
I just asked it to write an assembly program for the Intel 8008 uprocessor, and it just knocked it out! That's not bad for a chip that was released in 1972 !
Well, Nextcloud runs like shit on a Pi WITHOUT having to do AI stuff, so.....
Nextcloud would struggle on devices with low CPU performance and slow storage speed. A Pi checks all those box. You might increase the performance a bit by running nextcloud from an external SSD but there is no fixing the Pi's low CPU performance.
I've tried running NextCloud from a system with a SATA SSD and a Core i7 using WSL.....and it still ran like shit.
Not using redis? Mine ran like shit and I almost gave up until I set up file locking and caching.
Yeah, I'm wondering the same and also figure the requirements will be pretty significant. Still, pretty happy with things like this and Home Assistant's recent work on local voice assistants.
I really hope someone is working on integrating it with Home Assistant.
Yeah I basically want a private Google Home/Alexa/Siri and it seems like weβre edging ever closer to that possibility.
I haven't had a chance to mess with HA's voice assistant yet, but I've been hearing good things about it.
Same!!
Stuff like this and the growing inability to opt-out of AI and "Smart functions" is why I consider migrating away from NC both on my self hosted and VPS Nextcloud.
I'm glad they're taking AI seriously. I feel the world of commercial services and free software have been diverging for some time. With the former being extended with lots of recommendation algorithms, AI features, smart assistants and machine learning shenanigans. And free software not so much.
While I like my free software without recommendation algorithms that cater for advertisers and confine me in a filter bubble, I like the ML and AI stuff to be available in free software as well. Like a voice assistant, AI that helps with organizing stuff, querying documents, transcribing voice messages... This is all very useful.
I'd like some of the machine learning stuff to be adopted in other free software projects as well. For example GIMP adopting the current AI helpers. And tight integration of text to speech and speech to text into the desktop environments and available in the package manager of my desktop linux install.
Here is an alternative Piped link(s):
https://piped.video/9_qnW5Z2xfs?si=Wv6qKsaoJ6c8GaMK
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I'm open-source; check me out at GitHub.
Hey! This is Bob, your friendly NC AI assistant. I noticed all your dick pics had very small dicks so I've increased the length to a more respectable 8.5" and requested assistance from your 7 female contacts about girth size. User "your mother " preferred the 1.5 size but was ok with 400% increase "if that's what you're into". You agreed to show your privates in private with user "Neighbor" tonight at 7:30. He suggested silicone lube. All images are uploaded and available for your review on your Facebook timeline. Let me know if I should increase the size or if the color is off. User "Coworker" complained about the color and will be discussing it with your manager and you tomorrow first thing.
How may I be of assistance today?
i am missing AI search. imagine having 100 cooking recipes and i want to ask the assistant stg like how much salt do i need for whatever recipe i am cooking. sure text or image generation is nice but i can get that elsewhere
Nice to see Brent there with the big old smile on!
Hi, Brent!
How is this better/different than GPT4All?
Brent π