I've read that when they adapted the story of Desmond Doss into the movie "Hacksaw ridge" they had to tone down some of the true events, because the scriptwriters decided that audiences would find parts of the story unrealistic and unbelievable.
To be fair to Android, this is a limitation of the MTP protocol and not android.
To mount your storage as a mass storage device then the host device (your computer in this case) does raw sector read/writes to the device, the host device provides the filesystem services. For this to work it has to be completely unmounted from the phone as obviously having the block mounted in two filesystems at once would corrupt everything very badly.
Android used to split storage into various partitions, which is why this used to work - It was able to unmount the partition and let your PC manage it. This meant any apps using that partition needed to be stopped, etc etc. It was a pain, and I can totally understand why they moved away from this approach.
Personally I prefer the new way, yes using MTP has some limitations as you've noticed but it does mean the storage can remain mounted on android while your PC accesses it.
I can't think of any modern operating system that lets a foreign system mount it's already mounted filesystem over USB without going through some kind of server such as MTP, NFS etc.
In the case of plugging a "phone", as a device, into a USB host computer, the USB device (the phone) can present a filesystem endpoint to the host, and allow read/write access. The OS of the phone then passes these read/writes through to its mounted filesystem, with whatever mapping and access controls to the mounted filesystem it wishes to implement.
Thus the USB connection doesn't require that the raw filesystem of the phone be mounted by 2 hosts at the same time.
This already happens with every USB "stick" you plug into a host computer. The memory in the USB stick is accessed by firmware on a CPU inside the memory stick, which then presents that memory to the host as a USB storage class device. The firmware may not have a linux or iOS OS, but it does perform mapping to preserve and remap sectors to alleviate flash endurance issues, perform secure mounting, and other features.
There's no technical reason android can't do this.
p.s. MTP clients accessing an android device are a major PITA! Especially, ironically, for a linux OS USB host...
I think the issue comes with trying to present a USB mass storage API to the host where you allow the host to write to any arbitrary byte offset of the mass storage device without the structure that is imposed by a file system.
If the host and guest both get presented with the same “array of bytes” mass storage interface, then they will compete and potentially mess up each other’s reads and writes, let’s say if they both treat that array of bytes as an ext4 file system and try to write a file system metadata to the same physical location at the same time.
Of course you have have a “virtual file system” exposed over USB, but isn’t that exactly what MTP is? The point is that USB mass storage is not a virtual file system.
The problem is any of those things are effectively a reward for Russia for starting the war and invading Ukraine in the first place. Why should Russia get any advantage out of the war that they 100% started?? And pay them compensation! What a suggestion!
Russia is a bully. What do you think will happen if we have to pay the bully off each time they start smashing up their neighbors stuff up or just making threats?
And as for withdrawing NATO forces - NATO is a purely defensive organization. Its purpose is to defend against just the sort of shit Russia has pulled with Ukraine. If Ukraine was part of NATO the war would not have happened.
NATO is not a threat to Russia. Never has been, never will be. This is equivalent to a local crime lord complaining about being threatened by the police station down the road and demanding that the police station shuts down.
Are nuclear missiles located in Europe and pointed to the East also "purely defensive" weapon? It doesn't help good relations when you have a gun pointed at your face.
Yes, they are exactly that. The only (current) working deterrence/defensive strategy against an attack from nuclear weapons is the threat of a nuclear reprisal.
This has stopped a war directly between the major powers for the last 70 years and is known as MAD - Mutually assured destruction.
Its not a situation which anybody is comfortable with, but it works.
Honestly, this is basic cold war history stuff. Your question above shows you are either completely naïve or you consume way too much Russian propaganda.
Defensive weapon is something of an oxymoron, apart from technologies like missile defense [1]. Putting that to one side, rational deterrence theory[2] suggests that:
(Probability of deterrer carrying out deterrent threat × Costs if threat carried out) > (Probability of the attacker accomplishing the action × Benefits of the action)
You could argue that Russia successfully destabilising the US (via Trump) and Europe (via Brexit and far right) is proof that nuclear missiles "pointed to the east" worked at defending against direct conflict and forced an alternative.
I just tried this with a smaller "thinking" model (deepseek distill, running locally) and boy are you right. It keeps flipping between which direction it should turn, second guessing its thought process and then getting sidetracked with a different approach.
Not a true verification but I have tried the Deepseek R1 7b model running locally, it runs on my 6gb laptop GPU and the results are impressive.
Its obviously constrained by this hardware and this model size as it does some strange things sometimes and it is slow (30 secs to respond) but I've got it to do some impressive things that GPT4 struggles with or fails on.
Also of note I asked it about Taiwan and it parroted the official CCP line about Taiwan being part of China, without even the usual delay while it generated the result.
Not quite, I believe this sell off was caused by DeepSeek showing with their new model that the hardware demands of AI are not necessarily as high as everyone has assumed (as required by competing models).
I've tried their 7b model, running locally on a 6gb laptop GPU. Its not fast, but the results I've had have rivaled GPT4. Its impressive.
People who can use the 585B model will use the best model they can have. What DeepSeek really did was start an AI "space race" to AGI with China, and this race is running on Nvidia GPUs.
Some hobbyists will run the smaller model, but if you could, why not use the bigger & better one?
Model distillation has been a thing for over a decade, and LLM distillation has been widespread since 2023 [1].
There is nothing new in being able to leverage a bigger model to enrich smaller models. This is what people that don't understand the AI space got out of it, but it's clearly wrong.
OpenAI has smaller models too with o1 mini and o4 mini, and phi-1 has shown that distillation could make a model 10x smaller perform as well as a much bigger model. The issue with these models is that they can't generalize as well. Bigger models will always win at first, then you can specialize them.
Deepseek also showed that Nvidia GPUs could be more memory-efficient, which catapults Nvidia even further ahead of upcoming processors like Groq or AMD.
I believe you that it had to do with the selloff, but I believe that efficiency improvements are good news for NVIDIA: each card just got 20x more useful
That still means that that AI firms don't have to buy as many of Nvidia's chips, which is the whole thing that Nvidia's price was predicated on. FB, Google and Microsoft just had their their billions of dollars in Nvidia GPU capex blown out by $5M side-project. Tech firms are probably not going to be as generous shelling out whatever overinflated price Nvidia was asking for as they were a week ago.
Although there’s the Jevon’s Paradox possibility that more efficient AI will drive even more demand for AI chips because more uses will be found for them. But possibly not super high end NVDA chips but instead little Apple iPhone AI cores or smartwatch AI cores, etc.
Although not all commodities will work like fossil fuels did in Jevon’s Paradox. It could be the case that demand for AI doesn’t grow fast enough to keep demand for chips as high as it was, as efficiency improves.
> But possibly not super high end NVDA chips but instead little Apple iPhone AI cores or smartwatch AI cores, etc.
We tried that, though. NPUs are in all sorts of hardware, and it is entirely wasted silicon for most users, most of the time. They don't do LLM inference, they don't generate images, and they don't train models. Too weak to work, too specialized to be useful.
Nvidia "wins" by comparison because they don't specialize their hardware. The GPU is the NPU, and it's power scales with the size of GPU you own. The capability of a 0.75w NPU is rendered useless by the scale, capability and efficiency of a cluster of 600w dGPU clusters.
Wrong conclusion, IMO. This makes inference more cost effective which means self-hosting suddenly becomes more attractive to a wider share of the market.
GPUs will continue to be bought up as fast as fabs can spit them out.
The number of people interested in doing self-hosting for AI at the moment is a tiny, tiny percentage of enthusiast computer users, who indeed get to play with self-hosted LLMs on consumer hardware now.. but the promise of these AI companies is that LLMs will be the "next internet", or even the "next electricity" according to Sam Altman, all of which will run exclusively on Nvidia chips running in mega-datacenters, the promise of which was priced into Nvidia's share price as of last Friday. That appears on shaky ground now.
> That still means that that AI firms don't have to buy as many of Nvidia's chips
Couldn’t you say that about Blackwell as well? Blackwell is 25x more energy-efficient for generative AI tasks and offer up to 2.5x faster AI training performance overall.
The industry is compute starved and that makes totally sense.
The tranformer model on which current LLMs are based on are 8 years old. But why took it so much time to get to the LLMs only 2 years ago?
Simple, Nvidia first had to push the compute at scale strongly. Try training GPT4 on Voltas from 2017. Good luck with that!
Current LLMs are possible thanks to the compute Nvidia has provided in the past decade. You could technically use 20 year old CPUs for LLMs but you might need to connect a billion of them.
Always hilarious to see westerners concerned about privacy when it comes to China, yet not concerned at all about their own governments that know far more about you. Do they think some Chinese policeman is going to come to their door? Never heard of Snowden or the five eyes?
You can rent 10k H100 for 20 days with that money. Go and knock yourself out because that compute is probably higher than what DeepSeek received for that money. And that is public cloud pricing for single H100. I'm sure if you ask for 10k H100 you'll get them at half price so easily 40 days of training.
DeepSeek has fooled everyone by telling them that they need only so less money and people think that they only need to "buy" $5M worth of GPU but that's wrong. The money is the training costs of renting the GPU training hours.
Somebody had to install the 10k GPUs and that's paying $300M to Nvidia.
They only got more useful if the AI goldrush participants actually strike, well, gold. Otherwise it's not useful at all. Afaict it remains to be seen whether any of this AI stuff has actual commercial value. It's all just speculation predicated on thoughts and prayers.
When your business is selling a large number of cards to giant companies you don't want them to be 20x more useful because then people will buy fewer of them to do the same amount of work
each card is not 20x more useful lol. there's no evidence yet that the deepseek architecture would even yield a substantially (20x) more performant model with more compute.
if there's evidence to the contrary I'd love to see. in any case I don't think a h800 is even 20x better than a h100 anyway, so the 20x increase has to be wrong.
We need GPUs for inference, not just training. The Jevons Paradox suggests that reducing the cost per token will increase the overall demand for inference.
Also, everything we know about LLMs points to an entirely predictable correlation between training compute and performance.
Jevons paradox doesn't really suggest anything by itself. Jevons paradox is something that occurs in some instances of increased efficiency, but not all. I suppose the important question here is "What is the price elasticity of demand of inference?"
Personally, in the six months prior to the release of the deepseekv3 api, I'd made probably 100-200 api calls per month to llm services. In the past week I made 2.8 million api calls to dsv3.
Processing each english (word, part-of-speech, sense) triple in various ways. Generating (very silly) example sentences for each triple in various styles. Generating 'difficulty' ratings for each triple. Two examples:
High difficulty:
id = 37810
word = dendroid
pos = noun
sense = (mathematics) A connected continuum that is arcwise connected and hereditarily unicoherent.
elo = 2408.61936886416
sentence2 = The dendroid, that arboreal structure of the Real, emerges not as a mere geometric curiosity but as the very topology of desire, its branches both infinite and indivisible, a map of the unconscious where every detour is already inscribed in the unicoherence of the subject's jouissance.
Low difficulty:
id = 11910
word = bed
pos = noun
sense = A flat, soft piece of furniture designed for resting or sleeping.
elo = 447.32459484266
sentence2 = The city outside my window never closed its eyes, but I did, sinking into the cold embrace of a bed that smelled faintly of whiskey and regret.
the jevons paradox isn't about any particular product or company's product, so is irrelevant here. the relevant resource here is compute, which is already a commodity. secondly, even if it were about GPUs in particular, there's no evidence that nvidia would be able to sustain such high margins if fewer were necessary for equivalent performance. things are currently supply constrained, which gives nvidia price optionality.
> there's no evidence yet that the deepseek architecture would even yield a substantially more performant model with more compute.
It's supposed to. There was an info that the longer length of 'thinking' makes o3 model better than o1. I.e. at least at inference compute power still matters.
> It's supposed to. There was an info that the longer length of 'thinking' makes o3 model better than o1. I.e. at least at inference compute power still matters.
compute matters, but performance doesn't scale with compute from what I've heard about o3 vs o1.
you shouldn't take my word for it - go on the leaderboards and look at the top models from now, and then the top models from 2023 and look at the compute involved for both. there's obviously a huge increase, but it isn't proportional
Blackwell DC is $40k per piece and Digits is $3k per piece. So if 13x Digits are sold then it's the same turnover as a DC GPU for Nvidia. Yes, maybe lower margin but Nvidia can easily scale digits into masses compareds to Blackwell DC GPUs.
In the end, the winner is Nvidia because Nvidia doesn't care if DC GPU, Gaming GPU, Digits GPU, Jetson GPU is used for AI as long as Nvidia is used 98% of time for AI workloads. That is the world domination goal, simple as that.
And that's what Wallstreet doesn't get. Digits is 50% more turnover than the largest RTX GPU. On average gaming GPU turnover is probably around $500 per GPU. Nvidi probably sells 5 million gaming GPUs per quarter. Imagine they could reach such amounts of Digits. That would be $15b revenue and almost half of current DC revenue with Digits only.
Not quite, I believe this sell off was caused by Shockley showing with their "transistor" that the electricity demands of computers are not necessarily as high as everyone has assumed (as required by vacuum tubes).
Electricity demands will plummet when transistors take the place of vacuum tubes.
I've run their distilled 70B model and didn't come away too impressed -- feels similar to the existing base model it was trained on, which also rivaled GPT4
Exactly, and firing up reactors to train models just lost all its luster. Those standing before the Stargate will be bored with the whole thing by then end of the week.
that's a Milchmädchenrechnung. if it turns out that you can achieve status quo with 1% of the expected effort then that just mean you can achieve approximately 10 times the status quo (assuming O(exp)) with the established budget! and this race is a race to the sky (as opposed to the bottom) ... he who reaches AGI first takes the cake, buddy.
I'm assuming the parent poster is talking about using the Garmin Connect app, which does require connectivity. You are correct, the data is visible directly on the watch.
Its worth clarifying you are talking about the data on the phone app, which does require connectivity as nothing is stored on the phone app, its all on Garmin's servers.
However, most if not all of the data (recorded activities or health data) can be viewed directly on your watch, without any connectivity.
I have a lot of experience working with the Garmin API. The data you can see on the recording device (watch) is limited and basically worthless. Akin to looking at a raw csv full of data rather than nicely plotted over a map.
Sorry, hard disagree with everything you've said here.
I have a Fenix 6, I've worn it every day for the last 4.5 years. Its a brilliant smartwatch. I have multiple apps from Garmin's IQ store on it. Battery lasts between one and two weeks, thats including recording multiple activities each week, while using it to play music to my bluetooth headphones.
