belval

23 ms·

241.
by belval 3y ago
Unless these are special a "useful life" rating of 5000 cycles mean that after 5000 cycles your battery will be down to about 80% capacity compared to their original MWh rating. But full cycle is probably not the complete picture
242.
by belval 3y ago
Are these statements with any evidence? Especially those two bits: > when we permit people to do urbanism, even in America, they do it! > All the evidence suggests that typical American car-dependent suburbia is not the affirmative pr
243.
by belval 3y ago
I feel like the article misses the forest for the trees, yes kids like all those things, but they also enjoy having their own room, a backyard to play in and a street to play hockey in. The suburbs are part of a set of tradeoffs people make
244.
by belval 3y ago
There aren't a billion person on Earth to pay 20$/month. Not even sure that there are more than 100M. Paid services are for B2B these days and that's where their partnership with Microsoft knee caps any chances for astronomic
245.
by belval 3y ago
You seem to be seeing a confrontational message where I didn't mean one. Yes I'm sure you could find some old posts complaining about X that would illustrate the same phenomenon. My point is pretty much exactly that. Fast forward
246.
by belval 3y ago
You only get one chance to make a good first impression. Wayland is probably much better than when I tried it in 2019 but all I remember is the 2019 experience. I think a lot of people are in the same boat.
247.
by belval 3y ago
What a weird argument in the context of this bill. "If I'm too strict my child will resent me therefore I will vote so that my opinion can be enforced on everyone by the government and I can be my child's buddy". Must be
248.
by belval 3y ago
To be fair, Wireguard is incredibly easier to setup and maintain than OpenVPN, pretty much not comparable. I don't know how easy it is with Tailscale though so I can't comment as to how much harder Wireguard is compared to it.
249.
by belval 3y ago
As older researchers will often tell you: the job of a reviewer is to refuse your paper. It doesn't have to be completely black and white though. Maybe they did the maths and the results were middling/unclear, their calculations w
250.
by belval 3y ago
Yes, a lot of those have pretty egregious annotation mistakes. Once you get in high percentage it's often worth going through your dataset with your model prediction and compare. Obviously you can't do that on academic benchmarks
251.
by belval 3y ago
Don't look at absolute number, instead think of it in terms of relative improvement. DocVQA is a benchmark with a very strong SOTA. GPT-4 achieves 88.4, Gemini 90.9. It's only 2.5% increase, but a ~22% error reduction which is mas
252.
by belval 3y ago
> I find it naive to believe that the purpose of a for profit organization is anything but making profit. Where did I imply that Meta is doing it out of the kindness of their heart? Of course they want money and of course they think this
253.
by belval 3y ago
Your comment make it sounds like that's inherently negative. Google/OpenAI are drumming their message of "AI needs to be in the hands of the few" while Meta is actually releasing models and building a thriving open-sourc
254.
by belval 3y ago
> You cannot possibly review this in one minute. I didn't mean to imply that you can. It's more that at Amazon's scale, they are surely constrained by how many new items are put up for sale every second. If even just 1&#x
255.
by belval 3y ago
That part is confusing to me because it specifically calls out items that are for violating privacy. Any defense from "we are not responsible for how a product is used as long as it's legal" or "our safety team inspects
256.
by belval 3y ago
Those are all impressive numbers, but did they actually fix their software stack? Are there people here that have actually trained with those and if so how was the experience? I'm genuinely curious because it seems like every other AI
257.
by belval 3y ago
It's a bit of a mixed bag. NeuronSDK is implemented with Torch XLA so there's not real lock-in there. Your code will usually just work on CUDA if it works in Trainium. However for larger models (ZeRO, Sharding, etc...) they have t
258.
by belval 3y ago
It is probably a way to give workers fake kudos without actually giving them anything. A lot of north american modern management theory seems to think that praising your employees is the best way to get them to perform. By telling someone w
259.
by belval 3y ago
I have ~8 self-hosted services with docker-compose and one with snap (nextcloud). While I get that snap gets some flak for how cavalier it can be with your system, ultimately my nextcloud is always up to date and I've had very little
260.
by belval 3y ago
That gives me a new found respect for Mercedes-benz.
261.
by belval 3y ago
I love how people in the comments are just diregarding that simple fact. If you are "just" doing run-of-the-mill software engineering for a non-FAANG then sure look before you leap. As an employee of OpenAI, that went before congr
262.
by belval 3y ago
> We need to ensure that only appropriately trained users are sold these dangerous devices. If someone without a license wants to buy a tacoma and look at it sitting in their driveway what's the problem exactly? It's not like t
263.
by belval 3y ago
Why? Inferentia => inference, trainium => training. Given the usually naming of AWS product, having one where the name roughly matches what it does is pretty good? TPU is pretty good but is associated with Google. MTIA is an acronym b
264.
by belval 3y ago
> Amazon announced theirs last year Inferentia (inf1) was GA'ed in December 2019 so it's actually almost 4 years old now. The trainium (trn1) chips and the Inferentia 2 (inf2) refresh is indeed 1 year old though.
265.
by belval 3y ago
> My mother told me on my 12th birthday not to make the same mistake she did I don't want to read into your family situation at all, but this seems like an extremely weird thing to say to your pre-teen. Could she not wait for you to
266.
by belval 3y ago
I wonder where you live for that to be true, in my experience everyone just complains all the time about the previous guys' craftsmanship and that's especially true in trade and engineering. > restaurants KitchenConfidential is
267.
by belval 3y ago
Hot take: The advantages of WPA3 makes it irrelevant to most consumers and if RasPi saved a dollar by not including it then I am 100% ok with their decision. WEP was garbage because it could be cracked in minutes on a laptop. WPA3 is mostly
268.
by belval 3y ago
Ehh ITAR is very serious, national security level stuff. Sure it's not like agents will have you assassinated, but you will definitely be taken by men with guns (the police/FBI) if you try and export that tech.
269.
by belval 3y ago
For people wondering which of the links to click: https://news.ycombinator.com/item?id=36857676 It's mostly just the classic "nono if you don't agree it's because you don't understand" and &quo
270.
by belval 3y ago
> you have that problem with all heating and cooling This is probably regional, but in Canada it is common to have heatpumps/AC for cold and electrical heating/furnace for heat. That means your cold is coming from the top of th

More ›