I’d be fine if every vehicle had the hardware in it and it was activated via an upsell at the dealership. I think the truly insidious thing here is the subscription aspect, for all the various reasons other people in the responses have stated.
I’m ok with Intel doing the same thing. But if Intel started selling AVX512 with a monthly subscription I’d ever buy an Intel chip again.
You can have what I would consider deep knowledge of a system without the ability to manufacture it or modify it. For instance, we have pretty deep knowledge of how the sun or other stars work, but we can’t even begin to dream about creating one, or controlling one.
In the same way, we know a lot of how biology works. Obviously nowhere near all of it; but we are far beyond just scratching the surface. It just turns out that modifying a working complex system is pretty hard.
The overview didn’t give too many details so hopefully I’m not arguing against too much of a straw man, but I don’t find the idea that certain physical processes mean we are fundamentally different is sound. By their very nature, those are processes that can be studied and replicated either in hardware, wetware, or pure simulation. If it’s “randomness” from quantum mechanics, then we introduce an RNG. If it’s the process of wave function collapse, then we use a quantum computer. Etc, etc.
The OP said the service adds no value to the customer and I was specifically refuting that claim. Everyone knows that most addictive things are ok in moderation (I say most because I would argue things like cigarettes and certain other drugs have no benefits to society or the individuals who use them).
There is nothing wrong with food delivery when used in moderation. I use DoorDash once a week to get a lunch delivered on a day where I have meetings running through lunchtime. It saves me time and hassle, and is a fun change of pace. It also came in very handy when a family member of mine was very ill and I wasn’t in the same city. I could order food and have it delivered to them. The service is nothing like cigarettes, where the only point of the product is to get you addicted enough to keep buying more, and eventually give you cancer.
That’s not to say that DoorDash et al. aren’t enabling people with food addiction problems, just to point out that there are very many real good uses for the service.
Also depends on the complexity of the algorithm (specifically thinking of large neural networks). We have a model that requires 8 A100s for training due to the size of the activations. No way to replicate that on a local machine and have it train successfully in any reasonable time frame.
However the complexity of the algorithm many times scales with the size of the dataset, either the full corpus or the size of individual examples.
As someone who has used both, I whole-heartedly believe that anyone who would put some marginal financial benefits of a CU over the ease of use, consistency and convenience of a big corporate bank is insane. So it appears we’re at an impasse.
Any sane company does not want to be related to an organization where grooming, sex trafficking and rape are encouraged and practiced, I would think. That seems like more than enough reason to me.
I am no expert just a curious outsider, so take this with a large grain of salt, but it is my understanding that that’s one of the most pernicious misconceptions even in practicing psychologists, but that the current high quality research suggests the learning styles theory is flawed at best and wrong at worst. This article is ~8 years old but I don’t think anything has quantitatively changed the conclusions over the intervening years.
Proof of work to rank pages? That sounds like the worst idea I’ve ever heard. I’m no fan of Google injecting ads into the top spots of their results but at least you can get to the “real” results by just scrolling down. With this paradigm, every link becomes paid “ad”, and the ranking is completely dominated by companies and individuals with big pockets, not any objective or even subjective measure of the quality. And they’re all ads paid for by wasting energy.
It shows the history of the bill including votes. Each state does this differently, as well as the federal government. As for what happened to this bill, it never made it to a full senate vote, it died in a sub-committee, essentially.
I don’t think the general public sees entrepreneurs as inherently rational, but many entrepreneurs, especially Silicon Valley types, certainly think that of themselves.
Honestly, what a terrible letter. Did not once say the words layoff or let go. Using phrases like “continue to be part of klarna’s journey”. Good lord, is this a business or a cult? And not a single line about concrete steps they will take to support the layoffs (multi-month severance packages, etc). Hopefully that is what they intend, but as of right now, this is pretty disappointing.
Im an outsider to data center operations, but I work in machine learning. My suspicion is that this will be driven by either the next-Gen or next-to-next-Gen GPUs/AI accelerators. It’s the curse of the field that no matter how fast the chips get, and how much memory bandwidth they get, it’s never fast enough for the state-of-the-art research. So it’s probably naturally going to require some form of liquid cooling as the chips continue to get bigger and faster. And companies that build their business around machine learning will need to adopt these deployment patterns to stay on the cutting edge.
On the contrary, I believe MacBooks have continued to eat general market share since they switched to Apple Silicon, and they are by far the best built UNIX-based laptop you can get, and have by far the best battery life. There’s a reason so many devs (and scientists, like myself) use them.
This is targeting AMD GPUs and M1 GPUs currently not targeting the integrated Intel GPUs present in Intel machines. However if you have a 16 inch Intel MBP, or a Mac Pro, etc, this should work with your AMD GPUs. That support isn’t in the nightly packages yet (only Apple Silicon support so far) but the PyTorch team is saying that it will be available by the end of the week hopefully. If you just can’t wait, you should be able to build from source to test it out right now.
I think there’s two parts to this. The first is that there’s a minimum number of parameters required to do an arbitrary task. For any sufficiently complex task (image recognition, large language models, etc), it’s not clear how to find that lower bound. And the bound probably depends on the model chosen. But we do know that the more parameters you add, the more complex of a function you can learn.
On the other hand, for many reasons (energy, training/inference deployment complexity, latency, even some sense of model elegance I suppose), we don’t want to massively increase the number of parameters unnecessarily. But again, I don’t think we have great methods to estimate the “ideal” minimum number of parameters for a model to achieve its goals. And what we keep finding is that if you increase the number of parameters, and you increase the training corpus, the model gets more accurate, more impressive, and it’s not stopping.
So while I definitely agree that size for size’s sake is wasteful, I also don’t think we necessarily even know how to define “wasteful” for things like large language models right now.
It explicitly says in the article that the reason they capped it at ~1B parameters is because that’s the current limit of what they can achieve and hit their latency requirements? It has nothing to do with not being interested in scaling it further, as far as I’m aware.
Cool trucks, but if one of the concerns is weight, I can’t imagine a BEV truck with any decent range is gonna have appreciably better weight/axle. But perhaps I’m overestimating how much battery they’d need.
That sounds to me more like a team failing to adjust to a good WFH setup than anything intrinsic to the difference between WFH/office work. To be clear I’m not in an engineering role, I’m more of an applied scientist, but I started my newest job the day my company forced everyone to work from home for the pandemic, and overall my experience was very positive. Got lots of support and help, spent lots of time in slack/webex with senior members of the team, and overall had a great experience with onboarding. Only part I didn’t enjoy was IT working through all the hiccups of switching to WFH.
1) Unfortunately yes, I have seen some of that. I mean both the companies and the sideline quarterbacks.
2) Of course, but I’d hope most companies would be smart enough to be paying for productive output, not for bodily location.
3) Agreed. I personally prefer remote work, but I know many people who prefer going in - either to better structure their day, or to see others, and they should be able to do that. And I believe a competent company structure can support both of those work patterns. (At companies large enough for that sort of complexity)
I get people who want to work from home. I get people who want to work from the office. I get people who want to do hybrid. The one group of people I don’t get are the ones who suggest people should be forced to work from the office, and not able to decide for themselves. (Obviously it depends on the role - if you’re doing hardware design you need to go in sometimes, of course.)
I see some people in the macrumors comments saying “good riddance” like he’s being ungrateful or something. He’s a manager who’s standing up for what he believes is best for his team, and when he was overruled he did the only thing he could do to protest - quit. Good for him, and hopefully others follow suite. And as someone who works in the same field, he is absolutely right. ML teams right now are absolutely starved for good talent. Apple should be doing everything it can to hire and retain these people, including letting them live where they want.
It’s kind of crazy to think that Alex-net was just ~10 years ago, and look at what DALL-E can do now. I can’t imagine what the ML landscape is going to look like in another 10-20 years. Excited to be working in the field to experience it all first hand.
The statement about particle colliders is not really correct. The relativistic heavy ion collider, where I used to work, is about 2.4 miles in circumference, and reaches temperatures in the 10s of trillions of degrees in its collisions. This is > 1000 times hotter than a supernova. The environment in these collisions is made to reproduce the state of the universe fractions of a second after the Big Bang, and aren’t really comparable to any normal stellar activity.
A lab is a place to experiment. The person conducting an experiment does not need to be pushing the envelope of human knowledge to enjoy and benefit from the process.
I work in nuclear physics and no one of our experiments could be designed by a single person. They’re far too large and complicated, and even relatively simple sub-parts are multi-year, multi-million dollar undertakings.
Furthermore, even the ideas of “let’s use X to measure Y” are VERY rarely completely new and unheard of ideas from a single individual. Far more common that those ideas come from long term collaboration/discussions between multiple field experts.
I assume you did no research on her before commenting - she’s a very successful computer scientist and spent (a little) time as a VP at Google Cloud. While I don’t know if her background makes her the most qualified to be on a board, she is most certainly not a socialite or completely useless.