atty

I’ve never had to use CGAL directly, but indirectly it played a role in my PhD thesis, as a backend for a clustering problem called jet finding in high energy nuclear physics :) always get a little ping of happiness when I see these tools pop up.

As someone who worked at Kroger ~10 years ago, even then credit cards were probably only 60-70% of payments. A lot of cash and checks. The loyalty programs had their start when it wasn’t quite so easy to track purely via payment. Also helps to connect a person with a definite address to mail brochures/coupons to, and to link accounts when someone changes credit card numbers, etc.

When dealing with the Schrödinger equation, the eigenvalues are the energy levels of the quantum system.

I find it interesting how many articles and people on HN say something similar. I had basically the opposite thoughts (and outcomes) from my PhD. I loved the 5 years I spent in my program, I learned a ton about super interesting topics, I learned a ton of very applicable skills, I met and befriended many extremely intelligent and kind individuals, and I immediately got a job after graduating. I’m still at that job, and it’s amazing, 2 and a half years later. (And I didn’t do my work at an Ivy League or anything, I was at Wayne State in Detroit).

Frankly, the only thing I wish I would have realized beforehand is that you’re basically forming your young adult friend group with a bunch of people who are going to be spread out over the globe in ~5 years. That part is hard. Everything else was great.

I think there are many factors. First, the Chinese healthcare system is much smaller than western healthcare systems. Combine that with a much larger population and a pandemic and you end up in a bad situation really fast. Secondly, they’ve refused the mRNA vaccines, which are significantly more effective than their home grown vaccine. Third, and possibly most importantly, Xi et al both inflated the danger of Covid in their communications to the public, and made zero Covid a point of national and personal pride for Xi. He would lose face if they turned back now.

I am an American and not knowledgeable about how energy pricing works, so I am curious, how do we square this with the rates that Europeans have been paying for energy and natural gas lately? Are utilities making off like bandits? Or were they running very low over the last few months and suddenly got a huge influx?

Awesome thanks for pointing this out!

Sorry, on my phone at the moment so I don’t think I can really type some decent code right now!

I actually think that torch’s ShardedTensor looks very promising. Essentially you can initialize a sharded tensor from an already initialized tensor, or initialize a sharded tensor on a meta device where it’s not allocated locally and each shard gets initialized on the specified remote devices (useful for extremely large tensors)

The sharding is described by a ShardingSpec, where you can either let it shard equally sized shards across the requested devices, where the split happens along a single dimension, or you can do grid sharding along multiple dimensions. They also have a more general sharding spec that allows you to choose explicitly which indices go on which devices, if you need non uniform shards.

I think once these are implemented (along with some special cases like cloned tensors, and things like that), and once the distributed autograd engine has full support for CUDA, it should be pretty easy to start building out distributed versions of common neural net operations.

The one thing (that I haven’t thought about a ton, to be frank, and I’m sure other smarter people have :)) is that you’ll end up in cases with both a sharding spec for the weights as well as for the inputs, and what’s the best way to make sure everything matches up. Is the best way to handle that custom logic for each operation? And do you have each operation just reshard the input automatically? Seems potentially like a pretty big performance pitfall.

Just looking at commit frequency in the main branch, and activity on the relevant RFCs in GitHub issues. I have no visibility into what’s going on at Meta of course so it could just be there’s a lot of currently internal development, maybe there’s a lot of work going into a different branch at the moment, or maybe they’re waiting for things to happen in torch’s RPC support, I’m not sure. Or it could be they’re just waiting to release what they have as a beta in the 1.13 or 1.14 release?

I’ll note that checking right now there was a commit 15 hours ago, but the last commit before that seems to be 28 days. So some work is still going on at least, thankfully :)

Maybe “TF 3” or whatever they call it will be ergonomic and a pleasure to use, but that was the promise of TF 2, and unless you wanted to use Keras it was anything but. I’m glad I have been able to work only in PyTorch and Jax. Maybe TF will get better but I’m not holding my breath. On the other hand, XLA is very nice, and I hope they continue to develop it.

I hope torch has a convincing distributed tensor API coming soon. Their development on ShardedTensor seems to have slowed or stopped recently, so TF’s DTensor is definitely ahead, which is a shame. And of course TF’s ecosystem for the whole lifecycle is more mature with TFX, TF.js, etc, but torch is slowly closing those gaps, and hopefully that will continue.

Every browser on iOS is essentially safari but reskinned. Apple forces browser apps to be WebKit based. That means supporting other codecs, and different types of web technologies, etc, only happens on iOS when WebKit supports it, because all browsers have to use the WebKit that’s provided by apple.

If you’re ok spending 1 minute turning off their weird crypto-powered ad stuff (or if you’re ok with it), I personally think Brave is the best Chromium browser available, and I believe they have said they will not be removing Manifest v2 support. Also their built in ad/tracker blocking is nice.

What about this is healthy exactly?

They keep on pushing on the power of the iPads. Out of curiosity, does anyone really use their iPads for something they’d consider really compute intensive? I find that the best use of a tablet is reading and watching videos. Any time I want to do anything complex or computationally intensive I find a laptop to be much more efficient, both in terms of the OS flexibility and better input devices.

I work for an automotive OEM, and when a car has a serious recall, I don’t get to say “gee it’s frustrating to see the whole of my company get smeared for one engineering mistake I wasn’t a part of.” There’s no reason for customers or the public to try and figure out exactly what part of the company failed. It’s a systemic failure of the entire company, and it reflects negatively on all of us.

Same for you. You either need to fix Google’s long term support issues from the inside, or expect more of the same.

It can’t stop me from accessing index 1 when I meant 0 on a 3d array, so I think the claim doesn’t make much sense. Rust can stop me from accessing index N+1 though, which is useful for security. Of course almost every language also has support for bounds checked arrays either in their core or their standard library, and some for of array iterators, so I don’t think that’s a particularly compelling reason to choose Rust.

I worked at a Kroger (grocery store chain) for 7 years, and each time they brought in consultants to modify workflows they made our lives (stockers and baggers) worse, with no measurable improvement in efficiency. Needless to say, I am not a fan of consultants. Only time you should use them is when you’re doing something you have zero competency in. A grocery store should not be using them to design their stocking workflows.

Crap like this makes me wish RISC-V was the dominant ISA. Unfortunately it seems like the big players really don’t have any incentive to switch once they get their architecture licenses, since the ARM ecosystem and toolchains are so much more mature. I know Intel is working to support RISC-V in their foundry play, I hope that takes off in a big way.

Not that I feel particularly bad for Qualcomm here, they’re about as litigious, if not more so, than Arm.

I see a big mix of positive and negative in the comments, I’m gonna take a slightly different tact, and say I’m extremely bullish on the GPU market in the near-to-mid term on the consumer side. Obviously, a coming recession and relaxing of sales due to pandemic/lockdown pull ahead will probably dampen sales, but having 3 vendors taking 3 separate approaches to their designs is great for competition and consumer choice. We have Nvidia at the high end with monolithic chips probably delivering the highest performance in most/all areas, especially gaming. AMD is going with a chiplet architecture that should allow them to stay in the same ballpark of (non-raytracing) performance at a lower price. And finally we have Intel coming in at the low end with cards that seem like (once all the driver issues are sorted out) are going to be adequate for gaming, maybe even offer best in class price/performance in certain cases, and that seem to be also targeting productivity as well. They were the first out of the gate with AV1 hardware support, as well what looks like some relatively impressive matrix math units, similar to Nvidias tensor cores.

Overall looking very promising for the next 5-10 years, and Im hoping any potential recession doesn’t cause Intel or AMD to take their foot off the pedal, or in the worst case (more likely for Intel), even scrap the product lines.

From investor calls (where they actually can’t lie, at least not without fines and losing their jobs), they’ve had test samples of meteor lake for awhile already, which uses Intel 4. They’ve even had successful power-on samples for their 20A node (maybe also 18A?). It sounds like Intel is finally making some advancements on their process roadmap. Of course we’ll have to wait and see if they can hit acceptable yields, or if 4 becomes the next Intel 7 debacle. (Hopefully not)

Both AMD and Intel have released some truly impressive CPUs lately, and this month both of them are releasing their next-gen products (ryzen 7000 and raptor lake respectively, I believe). The new ryzen chips are looking very impressive, and if Intel’s announced numbers are to be believed, raptor lake is going to be pretty great.

Where they fall down is on perf/watt, not raw performance. However there’s a lot of things that go into that difference, not just ISA, so I’m not sure if anyone has really decided if x86 is fundamentally less efficient, or just currently less efficient due to current design choices and constraints.

This is pretty tangential, but my company just rolled out a new HR system. I don’t like the term Human Resources very much, because it implies a level of fungibity that isn’t really true, but the new one is simply astounding - they call it the “human capital management system”. I was awe struck that anyone thought calling employees “human capital” was at all acceptable, going from implied fungibility to full on “you are a replaceable cog in the machine”. I imagine the PM on that one got an MBA from a place that taught that management and labor need to be in a constant struggle because workers are greedy and lazy.

I was actually wondering if someone could explain where that cost increase is coming from. I know the design rules get more complicated as the process node shrinks, but I thought most of those design rules are essentially “taken care of”, because customers use building blocks from the foundry that already have those design rules baked in? And it’s still using the same software I thought?

It seems like a ton of these prefab projects show up on HN, but does anyone know of any that produce houses/structures that actually… look nice? It seems like everything I’ve seen on here have all the aesthetics of a shipping container. (Or, to be more on the nose, a concrete box).

I’m rooting for both Intel and Samsung. It sounds to me like Intel is ever so slowly getting back on track with Pat there (good interview on the Decoder podcast with him recently, if anyone is interested), but I’d guess we won’t really be able to see if Intel can get their crap together until ~2024-28, considering how long the lead time is for stuff like this. I don’t think Samsung had basically a failed decade like Intel did, but they haven’t been as quick or have as good yields on the last few nodes as TSMC has been, if I understand correctly.

TSMC being a node or 2 ahead has basically let them hold their customers by the balls, forcing them to take price increases and extreme lead times because there’s simply nowhere else to go for the same quality and performance. It would be great if we got to a place where we had 3 foundries on roughly equal footing, especially for national security and global supply chain robustness. But also because competition is good for us mere consumers as well, of course :)

The irony of the CEO’s t-shirt in that photo is… staggering.

I strongly suspect you’ve never been on the end of an internet doxxing/hate brigade if you can’t imagine how this could be used to make someone’s life a living hell.

I’ll explain again that I think they can be used for bad actions, and also that they should still be released, because the benefits will outweigh the negatives. It does not hurt to admit that some things can be dangerous when used in nefarious ways. No one suggests we ban kitchen knives even though they are lethal, because their utility is massive, and outweighs their danger. In much the same way these models have extreme utility, that almost certainly outweighs their potential negatives.

I believe you misunderstood my post. (Or I have misunderstood yours) I was not arguing that the models should not be released. I was pointing out that the statement that no harm has been done with them was false, and that even though that is the case it is probably better for society, on the whole, for them to be open.

We can both admit that the tools can and will be used for bad purposes, and come to the conclusion that their benefits outweigh the negatives. We would not be doing any favors to our own arguments by pretending otherwise.

There is a difference in ease of use. I could never use photoshop to fake something like that even if I wanted to.

Further, we have seen harm come from some of this already, there’s a pretty big online community that uses deepfakes to put people in situations they would rather not be in, the most obvious being porn.

There is a clear risk from these sorts of models as they get better - I mean recreating specific individuals’ likenesses in compromising images (or even worse, video). We’re not at that point yet, but these things are getting better fast, so it’s only a matter of time. The problem is that there’s no way to mitigate those risks except to keep the model behind an inference-only API, or not release it at all - as soon as the model is open sourced then fine-tuning the model can introduce whatever behavior you want. Holding the models back is virtue signaling at best and actively harmful at worst, because it draws attention from individuals who will take it as a challenge to find ways to misuse them, and cuts businesses and the open source community off from models that would otherwise be very useful to them, that they could help to improve. I’m concerned this is becoming a self fulfilling prophecy, where more companies will start to do the same thing, primarily because it’s what everyone else is doing.