JBits

At least you're working with Rust now.

As someone with zero knowledge of psychology, I'm biased against it. Partly because of my vague impression that psychology tries to fit people to models, rather than viewing models as limited approximations.

For a while I've thought it would be nice to know what results the field of psychology actually has that are trusted. Was there anything at all in the taught content which you liked? I didn't realise the DSM-V was that bad. If research on trans people can be cherry-picked, then does that mean that some reliable research exists?

My impression about reality is the opposite. The quantum world makes perfect sense while it's the emergence of the classical world which is unfathomable. The crazy "pop into existence" part is still incomprehensible, so I guess it's essentially the same.

I don't see how that changes anything. By this logic, there's no need for CoT reasoning at all, as a single pass should be sufficient. I don't see how that proves that CoT increases capabilities.

I would assuming that training a LLM would be unfeasible for a small research lab, so isn't tackling small problems like this unavoidable? Given that current LLMs have clear limitations, I can't think of anything better than developing beter architectures on small test cases, then a company can try scaling it later.

> CoT models can, in principle, solve _any_ complex task.

What is the justification for this? Is there a mathematical proof? To me, CoT seems like a hack to work around the severe limitations of current LLMs.

Some people just like em dashes—myself included. You can find em dashes in articles written by the author before LLMs became a thing.

The issue is that many of her videos argue that funding for particle physics should instead go into foundations and interpretations of quantum mechanics, specifically research completely identical to what she works on.

This is not helped by the fact that she pushes an interpretation of quantum mechanics viewed as fringe at best. Her takes on modern physics seem typically disingenuous or biased.

There are multiple competing quantum computing hardwares, so you haven't given all of them the same length of time.

What sort of tech/techniques did wondermoss use? Was it generating polygons?

Looking forward to the API when it's stabilised. Have there been any updates on the progress of allocators of this general area of Rust over the past year?

It matters, but it's not functionally equivalent between different architectures.

Since noone has many qubits, typically physical qubits are compared as opposed to virtual qubits (the error corrected ones).

The other key figures of merit are the 1-qubit and 2-qubit gate fidelities (basically the success rates). The 2-qubit gate is typically more difficult and has a lower fidelity, so people often compare qubits by looking only at the 2-qubit gate fidelity. Every 9 added to the 2-qubit gate fidelity is expected to roughly decrease the ratio of physical to virtual qubits by an order of magnitude.

In architectures where qubits are fixed in place and can only talk to their nearest neighbours, moving information around requires swap gates which are made up of the elementary 1 and 2-qubit gates. Some architectures have mobile qubits and all-to-all connectivity, so their proponents hope to avoid swap gates, considerably reducing the number of required 2-qubit gates required to run an algorithm, thus resulting in less errors to deal with.

Some companies, particularly ones on younger architectures, but perhaps with much better gate fidelities, argue that their scheme is better by virtue of being more "scalable" (having more potential in future).

It is expected that in the future, the overall clock speed of the quantum computer will matter, as the circuits we ultimately want to run are expected to be massively long. Since we're far away from the point where this matters, clock speed is uncommonly brought up.

In general, different architectures have different advantages. With different proponents having different beliefs of what matters, it was once described to me as each architecture having their own religion.

TL;DR: the two key stats are number of qubits and 2-qubit gate fidelity.

With examples ranging from Avatar the Last Airbender, to Castlevania, to Monty Oum (who also completely eschewed storyboards), I can't help but that reducing the over reliance on storyboards and focusing on the animation instead is a recurring theme among unusually successful animations/animators.

Flow's having an impact on animators seems great!

I believe they're migrating to Swift instead of Jaht.

What is the evidence that DeepSeek used OpenAI to train their model? Isn't this claim directly benefitting OpenAI as they can argue that any superior model requires their model?

There is less immigration to North England.

It's the only edition which is provided for free. This edition uses Lua 5.0, which is very similar to 5.1 which is still very popular as it's targeted by LuaJIT. The changes between 5.1 and 5.2 seem small as well.

I'm a bit confused as I thought muesli was healthy.

What did you need the eigenvalues for? I wouldn've guessed exact diagonalization but in that case you would need the eigenvectors.

I think plenty of companies would like to spend time training but aren't able to because the budget is too small. Still makes Sony reasonable for the shortage as you say.

It's not just that, it's also that training opportunities have become rare.

There really is an animator shortage in anime. There's a popular blog with multiple articles on it: https://blog.sakugabooru.com/2022/08/05/the-layout-crisis-th...

Though simply paying more sounds like it would fix the issue.

Fun and well written read. It's ironic though how terrible the 15% cut was considered given that Steam takes 30%.

Personally, I think it would've been better to keep Serenity OS support but drop the no 3rd party code rule. While I realise it would've gone against Serenity OS's write everything philosophy, it would've kept the browser as a rallying focus for the OS. Thinking about what actually happened, if the Serenity OS now treats Ladybird as a port, then I think the result will be the same minus the social effect.

Would you have any recommendations for someone looking at learning vdom frameworks today?

I think because Servo was orginally focused at researching new techniques for designing a modern browser, not creating something for users to user. Parts of Servo made it into Firefox.

If you're looking for something to do in Lua, I suggest making a game with the Love2D framework [love2d.org]!

Without knowing anything about CAD, it's surprising to me that the author didn't try any of these matrix solvers. Particularly since the case of non square matrices with iterative methods is common to many fields.

I think it's worth remembering that all of these AI's work by having an unbelievably large number of weights. So many weights that it's all an uncontrollable black box. On the other hand, your work is all about having control, and I don't personally expect your work to lose value for this reason.

Another thing to think about is what the AI is designed to do. Without knowing the details, I would expect it to be trained to produce the 'most likely' output given the prompt. Consequently, I would think being inventive is against its design, and 'most likely' is effectively that same as 'average'.

But it is solved theoretically, by quantum error correction. I'm not denying there are plenty of problems but can you justify that decoherence is the limiting problem for architectures other than superconducting qubits?

The application of linear algebra to rendering I can understand. Character animation less so, apart from the physics aspect.

Sounds like a nice field to work in! I didn't realise it used maths to that extent.

Would you have any recommendations for someone of a linear algebra / programming background if they were interested in this field?