HNHacker News
TopNewBestAskShowJobs

Tazerenix

200 karma · joined September 1, 2023

submissionscomments
Tazerenix··on UHZ1: NASA telescopes discover record-breaking black hole
One of two things:

Either what we know about black hole formation is basically complete (it goes gas -> star -> black hole -> accretion + collisions) but the environment in the early universe was sufficiently different/dense that parameters which rule out the formation of supermassive black holes now were different. Maybe there were many intermediate black holes just in the millions of years after the big bang and things were still close enough together that accretion could happen and collisions were "likely" at the rate needed to form SMBHs after just a billion years. If that is true we might expect to see many many active galactic nuclei as we get better telescopes and look further back, depending on how quickly such black holes formed.

The other option is there is a mechanism of black hole formation that bypasses the above chain which we understand. People talk about supermassive stars, gas clouds collapsing directly into black holes, or primordial black holes that existed due to essentially random distributions of density moments after the big bang causing some regions of space to collapse into massive black holes which then persisted. Such things are far more difficult to observe, but could be inferred if we don't see many many active nuclei as we get better telescopes but all other indications of the accuracy of the big bang + inflationary theory hold true.

Tazerenix··on UHZ1: NASA telescopes discover record-breaking black hole
Also important to note that the process of stellar collapse and then black hole accretion takes absolutely enormous amounts of time to collate a large amount of mass together. It's also an extremely energetic process, you would expect to see very bright black holes if millions of solar masses of matter were infalling creating very large and bright accretion disks. We do see some active galactic nuclei but not that many. There's just no way there was enough time for this to happen in the early universe, or really even after a measly 14 billion years (i.e. seeing these young supermassive black holes is challenging for the stellar collapse theory, but the theory was already pretty challenged).

Not to mention if supermassive black holes were being formed by accretion, you would expect to see many intermediate mass black holes (1000-1000000 solar masses) everywhere, but we see almost none.

Tazerenix··on Linear Algebra Done Right – 4th Edition
On that of course I agree, but mathematicians tend to "relegate" such things to exercises. This tends to look pretty bad to enthusiasts reading books because the key examples aren't explored in detail in the main text but actually those exercises become the foundation of learning for people taking a structured course, so its a bit of a disconnect when reading a book pdf. When you study such subjects in structured courses, 80%+ of your engagement with the subject will be in the form of exercises exploring exactly the sorts of things you mentioned.
Tazerenix··on Linear Algebra Done Right – 4th Edition
> I’m sure it’s important if you want to be a mathematician, but if you just want to understand enough to be able to use it?

This book is for people who want to be mathematicians.

Tazerenix··on Linear Algebra Done Right – 4th Edition
Mathematicians are well aware of complaints like these about introductions to their subjects, by the way.

It is for a reason that this book introduces the theory of abstract vector spaces and linear transformations, rather than relying on the crutch of intuition from Euclidean space. If you want to become a serious mathematician (and this is a book for such people, not for people looking for a gentle introduction to linear algebra for the purposes of applications) at some point it is necessary to rip the bandaid of unabstracted thinking off and engage seriously with abstraction as a tool.

It is an important and powerful skill to be presented with an abstract definition, only loosely related to concrete structures you have seen before, and work with it. In mathematics this begins with linear algebra, and then with abstract algebra, real analysis and topology, and eventually more advanced subjects like differential geometry.

It's difficult to explain to someone whose exposure to serious mathematics is mostly on the periphery that being exposed forcefully to this kind of thinking is a critical step to be able to make great leaps forward in the future. Brilliant developments of mathematics like, for example, the realisation that "space" is an intrinsic concept and geometry may be done without reference to an ambient Euclidean space begin with learning this kind of abstract thinking. It is easy to take for granted the fruits of this abstraction now, after the hard work has already been put in by others to develop it, and think that the best way to learn it is to return back to the concrete and avoid the abstract.

Tazerenix··on Artificial Intelligence and Copyright: Request for comments
Realistically an AI model is basically just a very complicated piece of software. The model weights are akin to the software code, the model outputs are akin to the outputs a user of the software creates, and the datasets are akin to the intellectual property put into the software by the developer to create the code.

In the same way that a developer could not simply steal someone elses intellectual property in order to develop a feature of a piece of software, one cannot simply steal the intellectual property to adjust the model weights. The main difference is its generally quite easy to see in practice if a model has utilized some intellectual property (because for example you can ask ChatGPT to recite the first 100 words of Harry Potter) compared to another piece of software where you'd need access to the source code or developers thoughts (which could only be achieved through litigation, in most circumstances).

I think a great many people come up with convoluted answers to this question because they are uncomfortable with the reality that these very large organizations have essentially stolen hoards of intellectual property, and now that the horse has bolted people want to justify not closing the barn door. It seems to me very simple: to train an AI model on data, you must respect its copyright. The model weights should be copywriteable by the developers of the model (even if the law currently does not allow this), and the outputs of the model should be copywriteable by the person who interacted with the model (software) to produce the outputs.

The analogy with Photoshop is extremely simple: If some other software invented Gaussian blurring and copywrited it, then Adobe would have to license that technology from them to include it as a feature in Photoshop. The actual photoshop software/code would be copywrited by Adobe, and if someone created an blurry image with Photoshop they can copywrite it.

I think people only disagree with this due to some sense that the process of translating data to model weights is "automatic" or "computational" in nature. You could in principle get a person to, by hand, go through millions of data sets and compute the changes to the model weights. This is no different to someone writing a piece of code, checking someone elses approach, and adjusting their own code after the fact. It just happens that we have developed very effective tooling to automate the adjusting of the code.

← PreviousPage 2 of 2