site banner
PaperclipPerfector  1d ago (text post)   1602 thread views

Small-Scale Question Sunday for May 10, 2026

Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?

This is your opportunity to ask questions. No question too simple or too silly.

Culture war topics are accepted, and proposals for a better intro post are appreciated.

1
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

I'm a firm believer that nothing important has happened in the PC space for the last 15 years

There is one exception to this: A gpu that can run local AI models. Not LLMs or image generators but the sorts of custom neural networks that have become quite common for eg. photo noise reduction, audio stem separation etc. In those cases even a fairly low end "good gpu" will do fine (such as this ancient NVidia Quadro P2000 Mobile aka Gefore GTX 1050) but the difference between "have" and "have not" is massive.

Interesting, does the software integrating those non-LLM AI functionalities offer to run this in the cloud for you? My experience with both image generation/manipulation and local LLMs has been that its almost always better to run those loads in the cloud - either directly from the big AI labs, through a vendor like openrouter.ai or on a rented GPU like runpod.io.

You can very well do it all locally, but it's a pointless toy with 8GB of VRAM, semi-interesting with 16GB and you're finally cloud-independent with 24 GB of VRAM. And you can get many hours of GPU time on runpod.io for the price you'd have to pay for this much VRAM.

But yeah, I haven't worked with audio models and the things Photoshop can throw at a GPU.

LLMs and other generation AIs are much more HW intensive than domain specific neural networks. 500M - 1B parameters is plenty when the model doesn't have to understand instructions or global context (hell, there are some task specific audio models in wide use that can fit in just a couple of megabytes while performing well). Sure, a more powerful gpu will run the models faster but when you're doing noise reduction on a dozen culled and selected photos it doesn't really matter if it takes 15 seconds or a minute to run the process when on cpu it would take half an hour. Likewise stem separation taking a few minutes is a non-issue when you only need to run it once or at most a few times for a song (such as when remixing or isolating instruments for practising the lines).

There are apps that can run in the cloud but they're (expensive) subscription or credit based and having to pay $30-$50 per month per app gets really expensive. Not to mention they tend to have ridiculous censorship (think photos of people on a beach where you want to remove some distractions and the cloud version complains that your content breaks the terms of use).

OK, all good arguments. Maybe I'll have to take a deep dive on what's possible in open source land on that front (I refuse to touch Adobe/Abelton et al.). Do you use commercial models/implementations or do you have a recommendation where to start if I want to set it up myself?

Not to mention they tend to have ridiculous censorship

That's the nice thing about just renting a VM on expensive hardware from runpod. They don't care at all what you do, because you provide all the software yourself.

Do you use commercial models/implementations or do you have a recommendation where to start if I want to set it up myself?

For photo noise reduction I've used Lightroom and (free but specific to Olympus / OM cameras) OM Workspace. I don't think there are good open source models as the models are camera specific due to different sensor and bayer filter characteristics (and why they can produce really impressive results - check the OM-1 original vs denoised comparison here where the original is far from what I'd consider usable while the Lightroom one is perfectly fine).

For audio stem separation Ultimate Vocal Remove is my tool of choice (you can download the models from the settings page). I start by removing vocals with MDX-Net / Kim Vocal 2, take the residual and remove drums with Demucs and then possibly remove the bass from the residual of that. Be aware that if you just split the stems from the original they will not sum to 100% and thus you want to go the recursive route. I'm sure there are some newer models that you can install manually but I haven't used those as the existing ones work well enough for my purpose (removing distracting vocals or emphasizing instrumental part to better hear what's happening).

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats