moonarchia
The Scientific Shitlord
- 28,720
- 55,352
She won't talk dirty to me, but she works well as an alarm clock and music box.
- 1
AI: The Rise of the Machines... Or Just a Lot of Overhyped Chatbots?
- Thread starter Sanrith Descartes
- Start date
Lanx
<Prior Amod>
- 73,722
- 174,408
the new ditzy voice lasted half a day b4 my wife reverted her back to the bland voice
also alexa+ is fucking chatty, had to tell her to shut up like x10
- 2
ToeMissile
Pronouns: zie/zhem/zer
- 3,664
- 2,463
Obsidian isn't an AI first tool. It's open source ( i believe) and sort of a blank slate with whole ecosystem built around it. There are a few note taking focus apps, the names escape me though.
If you're mainly doing notes, I would just use whatever has the best integration w/ your current notes/personal organization tool/office suite. For transcribing/notetaking, I mainly have a bit of experience with Copilot as it was available through my previous job. It's been ~ year since having used it, but it was pretty solid at transcribing and summarizing Teams meetings w/ 8+ people.
- 2
- 4,636
- 17,102
Thank you sir. That makes sense with Obsidian. My friend kept talking about in relation to it's AI compatibility, that I confounded the two. Thank you for the Copilot recommendation. I think I can get it with my office 365 account or MSDN subscription. I have 3 primary use cases. 1) Manage all my home labing notes and scripts. 2) documenting and commenting and setting todos while programming at work and aggregating my thoughts to writing proposals, reviews, policies, etc. at work.
Control
Golden Baronet of the Realm
- 5,283
- 14,971
I poked Obsidian a bit but never really got into using it. As Toe said, it's its own ecosystem with hundreds or thousands of plugins that can do pretty much anything. Complete overkill if you just have a straightforward task you want to accomplish, but it seems like it could be pretty great for life/project management or content management for writers/creatives. It kind of seems like wordpress for organization to me. It might be reasonable for your use cases if you want to have more control over your systems than whatever off-the-shelf things give you. It's free and can be run completely locally, which is nice. I know the people that love it, really love it.
- 4,636
- 17,102
Thanks. My buddy who recommended it got in during their early access phase and has a life time subscription. I kinda' wish I had joined when he told me about it. As I was looking through their publishing features, I started getting inspired. I have a ton of notes and scripts for my home lab including docker compose files and random useful proxmox commands sitting in random txt documents. It would be cool to publish a wiki for myself. I could also use it to teach my kids stuff. There are a lots of possibilities.
BrutulTM
Good, bad, I'm the guy with the gun.
- 15,570
- -181
I've been looking for a new notes app since Evernote got retarded. I messed around with Obsidian a bit but it seemed way too complicated. I tried Joplin for a while as well but it saved all the notes I transferred from Evernote as HTML which is dumb and they want money eventually too. I've kind of settled on Google Keep for now. It's free and it's easy to share notes with anyone with a Google account so I don't have to try to get family members to install an app and/or pay for it to share notes with them. It took me like 2 years to get my brother to actually use Evernote and the rest of the family is a lost cause. The only thing I don't like about Keep is that you can't attach pictures or diagrams in the body of the note. You can have them at the top, but if I'm at a class or a presentation I like to take notes and then put in photos of their PowerPoint slides embedded in the note which as far as I can tell you can't do on Keep. It is also kind of dumb about the order it displays notes in, but you can always find them with search.
- 2
Control
Golden Baronet of the Realm
- 5,283
- 14,971
You only need a subscription if you want them to host it for you. No problem to run it on you own machine locally or on your own server if you want online access.
- 1
- 1
ToeMissile
Pronouns: zie/zhem/zer
- 3,664
- 2,463
I use Notion a bit and like it better than Obsidian, a little more structure “out of the box” but still huge eco system. Don’t remember if you can self host, or you can hook up your own LLM account. Capacities is another app in the space
- 2
- 4,636
- 17,102
Haus
I am Big Balls!
- 17,913
- 72,869
So I have been tinkering with things and looking at how to build a model.
Obviously I don't have the data centers to train a model from scratch but I was looking at incremental retraining/additions to models. I've been looking at the concept of LoRA (which is small scale adjustment/retraining of models) and it's looking like this can be done on normal consumer grade hardware.
I have machines with a sufficient amount of RAM and CPU horsepower, would probably need to upgrade a GPU to something with 24G on it.
Has anybody else around here tinkered in this direction with things?
My first goal was just to take a model that was readily available and decently advanced that I could already run locally (leveraging something like GPT4all), and then start adjusting it to see what all I could change/modify about what it "thinks" and what it's "internal limiters" were enforcing.
Obviously I don't have the data centers to train a model from scratch but I was looking at incremental retraining/additions to models. I've been looking at the concept of LoRA (which is small scale adjustment/retraining of models) and it's looking like this can be done on normal consumer grade hardware.
I have machines with a sufficient amount of RAM and CPU horsepower, would probably need to upgrade a GPU to something with 24G on it.
Has anybody else around here tinkered in this direction with things?
My first goal was just to take a model that was readily available and decently advanced that I could already run locally (leveraging something like GPT4all), and then start adjusting it to see what all I could change/modify about what it "thinks" and what it's "internal limiters" were enforcing.
Control
Golden Baronet of the Realm
- 5,283
- 14,971
If you're talking about the image/video models (wan, qwen, etc), it's no problem to train loras on 24g, checkpoints/finetunes are similar I believe, but I haven't done that. I assume it's a similar process for non-visual models, but I don't know much about those. I'd guess that the resources needed to run a model are similar to the resources needed to train it though. How long it takes it another story, but you can always rent better hardware for a couple of bucks an hour if you really need. Training time depends on how much content you're adding and how many steps you're running of course, but for example, I think you should be able to train wan2.2 in 1-2 days, qwen in a few hours, z-image in an hour or so. I think you can train some of the older sd based models on very minimal hardware too, if you want to try out the process before upgrading. I used this guy's stuff as a guide: Ostris AI
- 1