Not to speak of managed cloud services that run on ARM under-the-hood/behind the scenes.
Of course ARM isn't inherently cheaper, AMD+Intel could cut prices/margins big and probably be competitive on $/perf
4,090 karma · joined December 23, 2020
Not to speak of managed cloud services that run on ARM under-the-hood/behind the scenes.
Of course ARM isn't inherently cheaper, AMD+Intel could cut prices/margins big and probably be competitive on $/perf
It's a highly competitive market. Companies aren't going to pay 100k/year to run a model on something that can run on a 2k consumer grade device.
128GB of gpu accessible/fast RAM can be had for $5000 on a macbook pro today. What will it be 3-4 years from now on linux/windows machines?
And we still haven't seen any SoC providers try to optimize for RAM capacity over compute yet.
Most of the rewards will be reaped by consumers rather than providers.
We're also in an age where the current levels of RAM in consumer devices were almost entirely optimized for prior to the existence of LLMs. I find it highly likely vendors will optimize for higher RAM capacity over other priorities in future hardware.
How long until a 256GB RAM laptop (shared with GPU) is reasonably cheap/available? I give it a few years at most.
It's possible that models grow orders of magnitude larger, but I find it more likely that the size of models will grow along the curve of cost of training decreasing/hardware cost improvements. There will be a sweet spot where it's economical to train larger models, and private companies won't push much beyond that.
You are comparing a 7nm processor to a 5nm one, and yet the gap isn't even very large. Which was entirely the OPs point.
Does a 5nm AMD chip perform similarly to the 5nm Apple chip at the same wattage? (Again, performance does not increase linearly with wattage, as you're likely to cite something violating this logic in the next response)
You seem to not understand the point being discussed though, so no reason to discuss further
What would your opinion be if they did?
Conceptually, why should a "donation" be any different than a "payment" in regards to the Apple tax? Regardless of their current legal agreement attempts to disambiguate these, they are conceptually quite the same. In one case you receive a good, in another you receive good-will.
Should Apple be entitled to 30% of these voluntary payments? Some of you will say "yes", but I'd bet pretty strongly the vast majority of the public would say "no".
And legislation will follow eventually, in my opinion, despite the US government/courts moving very slowly here.
Continued acts like these will only move up the timeline and increase the number of lawsuits which will eventually tip against Apple.
Why don't you supply your own sources? You're making a claim just the same as the OP, without providing any evidence in your favor. A good faith responder would do the legwork to provide a researched counterpoint.
To anybody that has actually been paying attention to CPU evolution over the years, the process node has clearly been the main differentiator between CPU performance. Intel had the process advantage and thus the CPU advantage, and now they don't.
Architecture matters too, but does not result anywhere near an order of magnitude difference, conventionally.
At least for my use case and when interfaced with through Kagi. Much higher hallucination rate.
GPT-4o hallucinates far less than Claude 3 Opus but also seems to have less niche knowledge (I was using it to assist with Groovy+Spock+Spring upgrade)
So I question a lot of the benchmarks published on the newest models. They don't seem to track linearly/accurately with my use cases
Gross federal tax revenue never declined, even in the year after the 2017 tax cuts took effect.
https://www.statista.com/statistics/200405/receipts-of-the-u...
There is no defensible moat unless a player truly develops some secret sauce on training. As of now seems that the most meaningful techniques are already widely known and understood.
The money will be made on compute and on applications of the base model (that are sufficiently novel/differentiated).
Investors will lose big on OpenAI and competitors (outside of greater fool approach)
They're in effectively everything, including fresh fruits, vegetables and meat. "Apples have an average of 195,500 plastic particles per gram"
I think this is one policy area most could agree on, and likely not an insurmountable cost
If we allowed Chinese autos into the country free of tariffs, prices would almost certainly decline materially in real terms over the next few years.
You can disallow bundling a video client with the video data provider, thus forcing the data provider to monetize by charging the clients to use the data. The clients make money either via subscriptions or ads, and selling new video data back to the provider.
e.g. Google would have to spin-off or re-org YouTube to split client/data and give same pricing terms to their client branch as to other third party clients
This is a lighter touch/market based solution, which I prefer to being overly prescriptive.
YouTube is evidence of that already.
A more straightforward way to accomplish this in areas where content size is large/expensive is to disallow coupling client creation with data hosting, and data vendors would license access to their data to client creators.
Bundling becomes anti-competitive at a certain network size, because there becomes no meaningful way to create a competitor network. The essence of what makes capitalism effective is competition driving costs lower, and in many areas in tech we have very little competition due to large network effects.
Keep in mind Capitalism != Free Market. A fully free market is a form of Capitalism that has no laws, and no impediment to monopoly formation. Competitive Capitalism with minimal laws to encourage competitive where large network effects or monopolies form is far more societally beneficial in the long run.
Decoupling client/data has already been done many times in the past in analogous situations, e.g. when movie producers were not allowed to own the theaters where the movies were played, giving a much more equal footing to smaller content producers.
For the data side, likely any requirements wouldn't go into effect until a dataset is deemed sufficiently large/societally important, and there could be a period of exclusivity similarly to the patent system to encourage innovation. This system works very well for new drug creation, with competitors free to copy the drug for pennies on the dollar after patent expiry, so I very much doubt it would stifle innovation in tech, especially given the lower capital requirements to innovate.
I'm not suggesting at all the government mandates private companies implement a public write api into their own datacenter. I'm suggesting the privately hosted data must be replicatable and thus hostable by competitors. Likely the practical way to do this, technically, is to support a public kafka/persistent eventing system such that anybody can firehose all historical and new data. Ideally with funding help.
Hosting data is cheaper than ever, and continues to deflate in cost. The companies in this line of fire are already quasi-monopoly behemoths, so I don't buy into the cost-prohibitive/stifling innovation perspective.
Client applications should compete on their individual merits, not coast on protocol lock-in.
Would WhatsApp or YouTube have as many users if others could build clients for the same data? (PII etc notwithstanding)
Protocols compete on the merits of the protocol, clients compete on the merits of the client.
I think this will be the reality/obvious a few decades down the line.
Unless you are directly, personally, targeted no hacker will waste the time trying to reverse engineer your algorithm... they'll just go on to brute forcing the next hash in the list.
And most people only have a few services that need to be truly secure anyway, which would use non-derived passwords (if they hack your netflix or spotify, who cares? Call support and get it back)
Password managers have had many exploits/failures over the years. You introduce so many points of failures bringing in a third party.
Your gibberish password with random symbols/characters isn't any more secure than a more memorable one of a similar length.
Only have to memorize 2-3 strings and more secure than a password manager since there's no third party in the loop.
Password Managers are a huge man-in-the-middle and liability in other regards (e.g. you don't have it present on a given device or on hand).
SSO from a single set of credentials is a much better solution. Multi-factor biometrics even better (outside of PII sensitivities)
Apple and Nvidia can switch to other fabs, but performance and competitiveness will be impacted. TSMC seems to be running more like a charity than a business with the amount of margin their customers get on their products (both Apple and Nvidia)
Theres a very recent comment from the CEO stating something similar... But historically it's seemed they've allowed their customers to take most of the margin.
They should have a lot of leverage being on the most advanced node, but doesn't seem they use it.
Likely a cultural thing, but if somebody has more insight would be curious
I tend to just relax during both boarding/de-boarding since it's only a few minutes difference anyway and the effort/stress of trying to move quickly outweighs the few minutes savings. If the seat is pre-selected, I can simply wait for the line to go away and board last without standing for 10m. It seems to me that most people rush to board first even with pre-selected seating (perhaps optimizing for space in the overhead bins)
If you have a connecting flight with a short layover, is the exception where I want to be right near the exit. So I guess it's more of a preference thing
Everyone has their own preferences, but I wonder is there a tangible advantage to flying Southwest to those that prefer it?
Agreed that this is largely a function of higher quality/longer lasting vehicles though
It's "too easy" to lose weight now.
It's like being upset at the discovery of toothpaste since it enables people to eat simple carbs and still retain their teeth, rather than relying on restraint and stoicism.
Or using insulin to manage diabetes rather than a no carb diet
In the macro sense there seems to be more widespread awareness and money going towards fab competition than ever before. I suspect we'll trend closer to performance parity between competitors in the coming years.
Nobody will be able to compete against Apple CPUs while they're 1 process node ahead of competitors though.
I recently moved to Linux from MacOS, but probably will have to go back to Apple on the m4 for local LLM capabilities (high RAM specs much cheaper than commercial GPUs). Too bad Asahi is still missing some pretty critical capabilities (microphone, USB-C monitor connection etc)
These are first mover prices.
What's the monthly health cost of being obese?
It's still expensive, but far cheaper than brand name with similar efficacy. Competition from these programs will likely drive costs much lower over the next few years.
Do people often seek out obscure pages without entering from a more common context?
e.g. hackernews to any article link