cschmid

Well technically, the starting points were always other elements like bismuth, and not lead. I believe the authors checked, and noted that in the paper: https://journals.aps.org/prc/abstract/10.1103/PhysRevC.111.0... )

I wonder what makes you say this? In my (limited) anecdotal experience, I've noticed that Austrians seem to have far higher levels of English competency than Germans, for example.

There's a song about this: 'When The War Came' by The Decemberists https://www.youtube.com/watch?v=XJHOiQ2uniU

For anyone who's wondering, you can see your current kernel config in a file like /boot/config or /proc/config.gz

Can I also interpret this as: 'AMD's pytorch support is so abysmal that inference is 10x slower than it should be'?

Not quite: half are on the left, and half are on the right of the vertical line; and half are on the bottom and half on the top of the horizontal line.

It's also wrong: SO(n) matrices have determinant +1.

Unfortunately, despite panels and batteries getting cheaper every year, not every rural household has the financial means to access these solutions. I'm really excited about initiatives creating microgrids with peer-to-peer payment systems, like solshare: https://www.youtube.com/watch?v=v33ERl42VYM

The only way to avoid falling under the provisions of the act is to not have any customers in the EU -- that's quite a big market to cut yourself out of, and definitely something a big company cannot afford to do.

My whole workflow of organizing and reading papers is centered on PDFs. While I like having interactive supplemental materials, I want to be able to print, save and annotate the papers I read.

Has anyone here found a link to the actual paper? If I click on 'paper', I only see what seems to be an awkward HTML version.

The Taiwanese scooter company Gogoro recently announced something similar: Their scooters come with swappable batteries that you rent. The swapping station they have throughout the city are basically walls full of batteries. Since the company owns all the batteries, they can then use them to provide power to the grid during peak load.

The article links to an old draft of the paper (it seems that the results in 4.1 couldn't be replicated). The arxiv has a more recent one: https://arxiv.org/abs/2212.13345

I disagree; I think a system like Wikipedia shows that a textual and visual format like markdown provides the flexibility to convey pretty much any information. For any formal and semantic storage system, you will probably spend more time adapting it to each specific use case than to actually fill it with content.

It's a popular science talk, not a presentation of a specific research result. The re-interpretation of scattering amplitudes in terms of novel mathematical structures is almost its own subfield of theoretical particle physics. Most of the specific results he mentioned are described in this paper: https://arxiv.org/abs/1312.2007

I don't exactly know what you mean, but at least most points in the blog post were covered in my undergrad numerics courses.

I'm not sure if this will actually be an HSR stop, but it would be super convenient for someone living in e.g. Fresno: Instead of having to use a connector flight to SFO, you can just take the train and transfer.

There are formal partnerships between rail providers and airlines around the world, where your train ride is treated like a leg of your flight. You have a guaranteed connection, and often your bag is transferred from the train to your airplane.

That would only help you if they claim to be US-based in the first place. They do mention background checks, there are some companies offering those internationally.

This is pretty obvious, no? If you choose a (uniformly) random point inside an n-sphere, the components of the vector will go to zero as n gets larger -- after all, the sum of their squares has to be less than 1.

A (uniformly) random point in an n-dimensional cube will have random coordinates from zero to 1, with no other constraint on their size.

Milpitas really is not walkable, although you can somewhat decently navigate it with a bike. Walkability mostly is a function of density, which you can just glean by eye from a satellite picture. Unfortunately, many otherwise pedestrian-friendly places in the Bay Area have homeless camps set up on them, which might be a deterrent.

Anything south of Oakland/SF is mostly strip mall hell, but parts of Alameda County, Contra Costa County, Marin County and San Francisco might be what you're looking for.

Sounds the same if you have a lisp.

I just tried replicating the same experiment using Jax's numpy API, and einsum is still slower, but at least the same order of magnitude:

  %timeit (x_jax @ y_jax).block_until_ready()
  579 µs ± 4.54 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

  %timeit jnp.einsum('bik,bkj->bij',x_jax,y_jax, optimize=True).block_until_ready()
  658 µs ± 1.38 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

  %timeit jnp.einsum('bik,bkj->bij',x_jax,y_jax).block_until_ready()
  660 µs ± 2.82 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

If I understand https://github.com/numpy/numpy/blob/v1.22.0/numpy/core/einsu... and https://github.com/numpy/numpy/blob/v1.22.0/numpy/core/src/m... correctly, using einsum without the optimize flag seems to use a for loop in C to do the multiplication.

The optimizer clearly tries to improve the performance, but in many cases, it doesn't seem to change anything. Let's simply multiply some matrices:

  x, y = np.random.rand(200, 200, 200), np.random.rand(200, 200, 200)

  %timeit x@y
  40.3 ms ± 2.52 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)

  %timeit np.einsum('bik,bkj->bij',x,y)
  1.53 s ± 21.8 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

  %timeit np.einsum('bik,bkj->bij',x,y, optimize=True)
  1.54 s ± 10.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

I'd really like to use einsum more often, because it allows me to code my expressions the same way I derive them on pen and paper. Unfortunately, as mentioned in the article, it's slow, because it converts your formula to a for loop.

So usually, I rewrite my formulas into messy combinations of broadcasts, transposes and array multiplications. Is there a package or an algorithm that does this conversion automatically? It seems to be a pretty straightforward problem, at least for most expressions I use.

No carrier should be able to brick a phone without it being locked to the network. I did some research online, and it seems that the culprit here is actually Best Buy (and Apple) for selling phones as unlocked although they actually aren't: Apple sells iPhones to third parties under a 'US Reseller Flex Policy' [1], which means that they are automatically locked to the first carrier they are activated on.

Nothing on the Best Buy website gives me any indication of this happening. If this is really true, it's deceptive advertisement. I'm not a lawyer, but this looks illegal.

[1] https://swappa.com/faq/answer/us-reseller-flex-policy

I don't know what the issue is. Let's assume for simplicity that the sensitivity is exactly 0.2 eV. Then if you measure something slightly above, like 0.201 eV, you can conclude it has a mass. If you measure something slightly below, 0.199 eV, you don't know.

I don't normally complain about this, but that result has been at least four times on the front page in the last two weeks:

https://news.ycombinator.com/item?id=29514642

https://news.ycombinator.com/item?id=29424749

https://news.ycombinator.com/item?id=29405380

https://news.ycombinator.com/item?id=29208141

Finding mathematical patterns and relations by fitting expressive functions has actually been a common technique in experimental mathematics (https://en.wikipedia.org/wiki/Experimental_mathematics) for a while.

In the last few years, neural networks in particular have been used a lot in areas of math related to string theory, where one tries to find structures on certain topological spaces that preserve or generate symmetries of the resulting physics. Here's a review that might be interesting:

https://arxiv.org/abs/2101.06317

Unfortunately, none of these mathematicians have the marketing prowess of a multi-billion dollar company...

I don't think replacing a workstation with a Macbook because of RAM makes too much sense: If running one minibatch of your model already takes up all the memory you have, where would the rest of your training data sit? In the M1, you don't have a separate main memory.

Also, software support for accelerated training on Apple hardware is extremely limited: Out of the main frameworks, only tensorflow seems to target it, and even there, the issues you'll face won't be high on the priority list.

I know that nvidia GPUs are very expensive, but if you're really serious about training a large model, the only alternative would be paying rent to Google.

I know this doesn't answer your question, but can I just ask (out of curiosity) why you're training ML models on a laptop?

I'm not sure I understand that logic -- Germany has far more potential customers than Austria. And it's not like the average German travels ten times as many miles by train as the average Austrian. Also, the BahnCard can only be used for some, but not all local transit options, which looks like a hassle.