Yes and just to add, the infra itself is pretty cheap. The cost comes from the labor and regulatory complexity. Budapest for instance has dirt cheap fibre just about everywhere.
When a telco provides poor quality service somewhere, people have no choice but to pay them as price takers. When there are options, telcos have to provide better service to win your business. Telcos with monopolies have always been rent-seekers. It happens time and time again that some newcomer comes up, and just the hint of competition gets Verizon/Spectrum/etc to suddenly build new tech and dig some trenches.
There is no one-size fits all solution here. It comes down to what the cost of spam/fake accounts is, the level of sophistication of your adversaries, and the cost of loss of use to legitimate users blocked by your signup gates. Each site has their own weighting across these factors.
Note that the (total) fertility rate normalizes away the population age distribution, and is often misleading. It does not tell you the actual increase/decrease in population. The US age pyramid is slightly more favorable than the Chinese one, and in particular China is about to face huge societal headwinds due to a large number of old people compared to young working age people.
Selective prosecution and tough punishment can still be a net positive vs no punishment. (I am not saying it necessarily is nor that I would celebrate this.)
The CCP derives a significant part of its legitimacy from improving quality of life and living standards for the common Han Chinese. Waste and fraud that harms consumers is a drag on this progress; the incentives somewhat align. Real economic harm often causes real political harm.
Right off the bat, I disagree with the assertion that software quality is merely a concept of how it functions now. In reality software is a living thing and quality is so much more than whether there is a glaring issue right now.
The airline industry gets huge subsidies in most countries to operate more rural/less profitable routes. Most of the passenger airports in the US for instance would not be viable without subsidies (most flights go to a few profitable hubs, but the long tail or airports forms the majority by count).
Amazon delivers everywhere because USPS subsidizes package delivery to unprofitable areas. You don’t get next day prime except in a relatively small proportion of the country (by area).
Suppose 98% users have not had any sessions crash. You want to build an addon feature that 10% of your users will buy and which will increase the revenue from those users by 30%.
Do you spend time building the feature, or trying to understand why 2% of users sometimes see crashes?
This should be divided into three parts: marketing and selling people questionable combo drugs at insane cost (bad), the case of oral phenylephrine (idiotic + bad), and the efficacy of the other drugs in the mix (guaifanesin, etc) (unclear).
This post seems quite far fetched. Amazon is well aware of the paradox of choice, and the vast majority of UI changes I have seen recently are exactly those that guide and reinforce you to buy one option, without the decision paralysis. Items are not homogeneous, and it is obvious that they try to concentrate purchases to a smaller set of SKUs to reap the same benefits as Costco. It’s simply that Amazon can additionally support the long tail of SKUs with a heterogeneous warehouse system (and heterogeneous profit margins).
On the delivery side: US suburbia is just in general not a sustainable solution. Delivery is just one way in which it bites. Somewhere like NYC, the amortized delivery cost (internalized or externalized) is very low (and opposite to Costcos which require a drive to an inconvenient location).
The bit about agents doing your shopping is falling for the same trap as crypto people thinking NFTs will kill Ticketmaster. These have never been technical problems: the APIs don’t exist for nontechnical reasons.
I thought this comment would go in a slightly different direction: the body of work that is mathematics has plenty of “bugs”; proofs with mistakes or other human errors. Yet we take the body to be correct (we believe it “works”) in aggregate, partly because the intuition of mathematicians tells us that these bugs are solvable and don’t bring down the whole. Of course the less intuitive/more surprising/more central the statement, the stricter the standard for proof and more eyes that have walked through it.
I am not sure which part you are interpreting as underestimation or whatever? Quite the opposite: I claim the difference arises from a difference in strategies, not from intrinsic differences in ability.
Also I was responding to a claim about what will happen in less than 6 months (that’s about the edge of what you can meaningfully say too much about in this field).
These strategies take materially different resources; it’s not an overnight decision made by leadership. I suppose there is a natural experiment ongoing at Meta regarding this, it seems they recently moved a number of people into a division to produce such data overnight. So we will find out soon how quick they climb the leaderboards.
I think you are making a distinction between pre training and later stages? The value on eg Fable output is exactly the careful preference optimization embedded in those responses. Not all data is the same (sorry if my first comment was sloppy on that).
The article makes a very specific claim with a clear deadline less than 6 months ago. I do not underestimate the Chinese labs and their capabilities, if they wish they can retool to start overtaking the US labs with a different strategy. My comment shouldn’t be read as a permanent impossibility statement, just an observation on where we are right now. At the moment their strategy seems to be to produce decent quality, highly optimized models; and a pivot will take longer than 6 months to materialize into overtaking the frontier labs (that themselves do not look like they will throw the towel in in the next 6 months).
The Chinese models will not overtake the frontier US ones given the current way things are going. The US models derive their lead from incredible efforts to source more and higher quality (mostly synthetic data) via great feats (eg generating with humongous teacher models that could never feasibly serve interactive traffic). The Chinese models advance via heroic efforts to optimize models and great feats to secure more and higher quality training data from the US frontier models.
For an (Chinese) open weight model to surpass the (US lab) frontier models, this equation must flip and the Chinese labs must entirely retool from harvesting frontier model data to producing the data systems and efforts to produce novel data; as well as procuring latest generation hardware en masse for this. This does not happen easily. Also training a frontier scale model is actually not such an unimaginable feat: doing all the inference with the teacher models is where the hardware goes.