> My limited understanding is that Nerfs are compute-heavy because each cloud point is essentially a small neural network
There's no point cloud in NeRFs. A NeRF scene is a continuous representation in a neural network, i.e. the scene is represented by neural network weights, but (unlike with 3D Gaussian Splatting) there's no explicit representation of any points. Nobody can tell you what any of the network weights represent, and there's no part of it that explicitly tells you "we have a point at location (x, y, z)". That's why 3D Gaussian Splatting is much easier to work with and create editing tools for.
So that would be runnable on a MBP with a M2 Max, but the context window must be quite small, I don’t really find anything under about 4096 that useful
That's a tricky number. Does it run on an 80GB GPU, does it auto-shave some parameters to fit in 79.99GB like any articifially "intelligent" piece of code would do, or does it give up like an unintelligent piece of code?
Are you asking if the framework automatically quantizes/prunes the model on the fly?
Or are you suggesting the LLM itself should realize it's too big to run, and prune/quantize itself? Your references to "intelligent" almost leads me to the conclusion that you think the LLM should prune itself. Not only is this a chicken and egg problem, but LLMs are statistical models, they aren't inherently self bootstraping.
I realize that, but I do think it's doable to bootstrap it on a cluster and teach itself to self-prune, and surprised nobody is actively working on this.
I hate software that complains (about dependencies, resources) when you try to run it and I think that should be one of the first use cases for LLMs to get L5 autonomous software installation and execution.
The LLM itself should realize it’s too big and only put the important parts on the gpu. If you’re asking questions about literature there’s no need to have all the params on the gpu, just tell it to put only the ones for literature on there.
Most of these self-help books are basically what happens if you take what could be a decent blog post and just blow up the word count until you can publish it as a book.
The Machine Learning community is still overwhelmingly on X, which likely explains your experience. There are other communities, like that of Ape/iOS developers, that have moved to Mastodon, and for which the quality of conversation is now much higher on Mastodon than on X.
Or just don't disclaim that your comment comes from GPT as there is no way to prove otherwise and it won't annoy any hater beyond the actual content quality.
If you don't think it's interesting you are free to skip it after they said it was a chatGPT answer. I don't really like you deciding for everyone on hackernews what "should" be posted.
This is a "comments" section. ChatGPT didn't crawl here, click "reply" and post its comment. Forwarding other people's/AI's words is not a `comment` and violates the spirit of what a comments section is.
I agree with you if the comment was a straight up copy and paste of what chatGPT said. But it wasn't. The comment provided context and explained why it has value. Just because something is generated by chatGPT doesn't mean it cannot contribute to a discussion.
See it like a cache, no need to recompute. Or like these nice archive links people post for paywalled articles, I could do it myself but it's simply helpful.
Same for me. I've been mostly vegan for over ten years now. I "downgrade" to only vegetarian when it really cannot be avoided – which, luckily, has been only when I've travelled to countries where there really weren't any vegan options.
> On the other hand, a lot of people who claim animals have thoughts and emotions seem to think that cows have complicated human-level thoughts like "I am an oppressed cog; my owner will send me to the glue factory when I am too old to give milk, and yet I must queue up regardless, for my spirit is broken; my calf has been taken and I will never know if he got a college degree; life is pure suffering." This seems unlikely to be true.
As a current compsci masters student at ETH Zurich: yes, having those explanations is wonderful, but that just means that topics like SVD and PCA are taken for granted now, and the things that we need to chew on are things for which no nice canonical explainers have been created (and that likely weren’t taught in 1990).
Personally I think that innovation in explanation/illustration of abstract mathematical topics is one of the most valuable kinds of progress, but it’s rarely talked about as such. When someone comes up with the right framework, visualisation, or metaphor for explaining a complex and abstract subject in a way that induces “the right” mental model, and this new teaching “tool” proliferates, that is just a beautiful thing to see.
There's no point cloud in NeRFs. A NeRF scene is a continuous representation in a neural network, i.e. the scene is represented by neural network weights, but (unlike with 3D Gaussian Splatting) there's no explicit representation of any points. Nobody can tell you what any of the network weights represent, and there's no part of it that explicitly tells you "we have a point at location (x, y, z)". That's why 3D Gaussian Splatting is much easier to work with and create editing tools for.