To correctly assess the state of the world. Since Tesla exclusively uses visual sensors, they are massively limited in how accurate it can ever be… or safe.
Evolution has no goals, not even survival. Evolution is something that happens. Some species survive for a while, others don't.
Think of it like saying water has the goal of flowing down the mountain along the path of least resistance. Of course it doesn't, it's just something that happens. There's no goal.
That is true, but the point is that having swap available, increases the time between recording samples, and needing to commit them to disk.
Well-written, long term recording software doesn’t quit or crash. It records what it needs to record, and - by using swap - gives itself plenty of time to flush the buffers using whatever techniques are necessary for safety.
Disclaimer: I’ve written this software, both with and without swap available in various embedded contexts, in real products. The answer to the question is that having swap means higher data rates can be attained before needing to sync.
> Well-written, long term recording software doesn’t quit or crash.
Power outages, hardware failures, and OS bugs happen to the finest application software.
I believe you from your experience that it can be useful to have recorded buffers swap out before flushing them to durable storage. But I do find it a bit surprising, since the swap system has to do the storage flush you are paying for the IO, why not do it durably?
The fine article argued that you can save engineer cycles by having the OS manage optimizing out-of-working set memory for you, but that isn’t what you’re arguing here.
I guess the point is, sometimes you just need a lot of memory and want to record into it as quickly as you can.
Then, when the time is right, flush it all to disk.
The VMM is pretty good at being tight and productive - so use it as the big fat buffer it is, and spawn worker threads to flush things at appropriate times.
If you don't have swap, you have to flush more often ...
They have already spent an enormous amount of money. It’s hard to see how they could make it back quickly, if ever. I’d like to be wrong, but I expect they will continue to be a money losing experiment for a long time yet.
How much money they've spent in the past is irrelevant. That money all came from investors, in exchange for a stake in the company. It never needs to be "paid back". Besides which, those investors have earned all those funds back already, and then some (on paper).
All that matters at this point is how much money they'll lose/earn in the future. There are no shortage of investors willing to put money into this effort, and they're growing exponentially, so there won't be any pressure for them to turn overall profitable for several more years.
Boeing may never make back the development costs of the 787. That was an absolutely epic disaster of a project. But that doesn't mean Boeing shouldn't build and sell every 787 they can profitably sell.
If Waymo is at breakeven including capex, opex, and overhead, operations logistics becomes the limiting factor. While Alphabet is capable of investing more money into Waymo, I think they've reached the tipping point. If you see Waymo expansion accelerate, bet on that tipping point having been reached.
Uber took 14 years to make it to profitability. Money's frequently characterized as impatient, unable to look past the next quarter, but when it wants to be, it can wait.
Waymo's older than Uber, but they hold many key patents by this point. Now that they've started running a taxi service, it seems straightforwards to scale up, assuming that is the business they want to be in. Then it's just a matter of charging more than it costs to run the service, and wait.
What makes investors patient when no profits for years, is when they see growth, entrenched commanding lead and network effects, large user base etc. As long as investors can imagine a good likelihood of eventual profitability, then growth in the present is a fantastic substitute for profits.
Growth tells you the eventual profits will be bigger. Leadership and moat gives certainty that the company will actually get the profits for the market they grew.
How much money do they make off the average person in the value of ads shown per year?
Now compare to how much money the average person spends on driving per year.
If Waymo winds up running half the market in autonomous transportation over the next several decades, it'll make search look like peanuts in comparison.
You need to consider profit margins. The cost of showing somebody an ad is very near $0, which is what makes digital products so profitable. But when you do things in the real world, especially in highly competitive markets where the customer is extremely price sensitive, your profit per mile is going to approach $0. For instance WalMart's profit per item sold is less than 3%, and for driving this will likely be substantially lower (given the combination of customer price sensitivity + competition). The way you make up for this is in massive volume, but Waymo for now remains a heavily ringfenced operation and so it's not entirely clear how they reach scale. Google also has a very poor record of long-term performance in competitive markets.
The winner in self driving will likely be enabled by extreme vertical integration - you want to be building your own cars, cleaning your own cars, repairing your own cars, and so on.
The average American sees something like $500 of ads that go through Google per year. There's a profit margin of around 50% since Google has to pay publishers and pay for running search. So that's $250/person in profit per year.
The average American spends something like $12,500 in car+taxi/rideshare per year. Suppose with Waymo that goes down to $7,000 and it's 20% profit. That's $1,400/person in profit per year.
Obviously it gets much more complicated -- the profit margin depends on whether there are serious competitors to Waymo and how much Waymo's head start matters. Waymo will bring costs down further with shared vans and buses on demand. Profitability will rise with video ads in vehicles that you pay not to see. And so forth.
But autonomous rideshare is going to be larger than search any way you look at it. Profits won't be as high as search, but the barriers to entry are so high that profits will be high for a long time.
Those data you referenced are per household, not per person, and the majority of that is loan+insurance. The actual cost in terms of maintenance, fuel, etc is quite low, and that's the price that eventually will be the goal line for autotaxi companies. 20% [net] profit margins do not generally exist in competitive real world industries, outside of perhaps something like real estate. A net profit margin of 5% would be huge, and I think it will likely be much closer to 1%, or even less, simply because in the end it's going to be a commodity where all that matters is price.
I also think you're overestimating the impact of things like ads, buses, etc. The second Waymos become less pleasant than any remotely comparably priced option, they will lose customers.
No it's per adult not per household. The average household has 2.2 cars, so the figure per household is much higher. And it doesn't matter what proportion is loan vs insurance vs maintenance vs fuel, because Waymo replaces literally all of it.
And yes I assume Waymo will have high profit margins for an extended period of time because they have such a massive head start, and for a long time will be competing primarily against rideshare with human drivers, so won't be pushed below that. Their marginal costs will be much cheaper than that, not having to pay drivers. Hence 20% is not unreasonable.
Then, even in the long term, the economies of scale they develop and network effects will continue to give them a significant advantage. Not 20% margins, but way more than 1%. Especially as they start to vertically integrate the hardware at some point.
Here is where you would generally cite sources. [1] Those are the data from the BLS. Total transport spending per household is $13,174. The term they use is consumer unit, which you may have conflated with consumer/person, but it's practically the same as household. There are 134m consumer units, and 131m households.
Waymo is currently charging substantially more than Lyft/Uber and is not profitable. Human drivers can taxi in anything with 4 wheels and a hood, and its 100% their responsibility to take care of their vehicle, fuel it, clean it, and so on. Each Waymo currently costs ~$200,000 and is going to have a proportionally higher maintenance costs, and all of those costs must be covered by Google. So their costs are far higher than you're ballparking.
As for competition - Tesla has already launched a live robotaxi trial in Austin, so it's already here.
I was just going off the top Google result based on AAA data. Took a closer look and it turns out it's the average for new cars [1], so the discrepancy must be that your statistic takes into account the secondhand market. Thanks for the correction.
In any case, the overall point is the same -- it's a vastly larger market than Search. And what Waymo currently charges, and the current cost of their cars, is irrelevant. Waymo's business model isn't based on the economics this year or next year. It's based on the economics ten and twenty years from now, when costs have fallen dramatically as they switch to cheaper models and gain massive economies of scale.
As for Tesla, it's hard to take seriously given all the promises it's made and completely failed to deliver on. Their trial currently has a safety human in a front seat and is limited to a tiny group of testers. It's so many years behind Waymo already, and it's unclear if the technological approach it's taking will ever be able to catch up or meet minimal safety requirements.
Can’t imagine Tesla will be able to remove the passenger seat safety monitor any time in the next 5 years. Refusal to install lidar means Tesla’s AI has to be 100% perfect, which won’t happen for a long time, if ever.
When these are ubiquitous enough, the vast majority of people who currently own cars won't need to. It'll be so much cheaper and easier to use rideshare.
I can't really imagine the circumstances where I wouldn't want to still own my own vehicle even if it had an autonomous mode. I drive it places where I don't have cellular service. I keep lots of stuff in the vehicle. It's customized with accessories like roofracks. I can hop in my vehicle from my house immediately whenever I want to.
If I lived in a city and garaging a car were inconvenient/expensive? Maybe. But that's not me or a lot of other people.
But if it's half the price over the course of a year? And you can summon it in advance cheaply? And it basically never takes more than 5 min to arrive anyways, since they're everywhere?
You might decide it's worth it to keep the stuff you really need in a messenger bag or backpack or something, the way people in NYC do. And maybe roof racks don't matter if you can just summon a second autonomous van behind you to hold whatever you were going to put on your roof.
Obviously if you're a contractor or something you'll need your own vehicle. But the point is that for most people, sure they can't keep stuff in their trunk all the time, but that's a happy tradeoff if the total cost of driving is 50% less.
Statistically, and from a global perspective, the apartment-dwelling car owner (most likely with a lower income than yours) is a heck of a lot more common than living in American-style suburbia or a small town.
Of course there will be exceptions or holdouts, but it will come for gig drivers, then for second cars, and go from there. There will be versions with roof racks, with extra luggage space, with child seats.
Imagine if you could buy your own "Waymo-equipped car". No need for driving lessons. No aggravation. No road rage.
How many people would pay for such a luxury car? With the US population aging and public transit non-existent in most places, Waymo probably has a market for cars.
There’s clearly a demand for self-driving privately owned vehicles as well, but think of it this way - why own a self-driving Chevy when you could hire a self-driving Cadillac when you need to go somewhere?
That wasn't the MPEG design goal. It was to stream video through a distribution network where dropouts would be tolerated as part of doing business. People were accustomed to snowy analog broadcast video. That is more disruptive when listening to purely audio. This is incidentally why CDs had their error handling significantly improved over Phillips' original prototype which would have been much more susceptible to scratches if commercialized.
My wife had Crohn's disease since a teenager, and was diagnosed with metastatic gallbladder cancer aged 52. A death sentence. She chose to do aggressive chemo to prolong her life from a few weeks to a few months. She suffered a great deal before she died.
An immunotherapy treatment was discussed, and it could possibly have helped a lot, but it carried a somewhat high risk of causing a disastrous Crohn's flare that would kill her immediately. The doctor was unwilling to try this because it might kill her. So she died inevitably without it.
It was a classic medical ethics case right there in our crisis. We did a lot of interesting and intense things in those months before she died. Fuck.
>An immunotherapy treatment was discussed, and it could possibly have helped a lot, but it carried a somewhat high risk of causing a disastrous Crohn's flare
I'm really sorry to hear about it playing out that way man. What a horrible dilemma to have to be in. Also, want to ask because I have a few people close to me with Crohns: Obviously there's a lot of nuance and detail in this kind of combination of two illnesses and a specific, very complex medicine, but would you mind sharing a bit more detail of why the immunotherapy was so risky for such a deadly flare? I know immunotherapies are sometimes used to even treat autoimmune diseases, so I'm very curious for this reason too.
The problem is that the cancerous cells are themselves descended from healthy cells; other than the small differences that make them cancerous, they are in fact the same thing.
Many of the cutting edge immunotherapies for cancer essentially teach the immune system to target the cancerous cells.
However, in combination with an autoimmune disease like Chrons where the immune system has already learned to react to healthy cells, there is a much higher chance that an immunotherapy intended to target only cancerous cells also causes the immune system to target more healthy cells.
Yes, exactly. The specific risk for Crohn's is that the amped up immune system will rapidly kill ALL the gut cells that they've been pissed about for decades.
For the unfamiliar: Crohn's is to guts what eczema is to skin. They are both autoimmune diseases where the immune system attacks a specific kind of healthy cell. Unhelpful.
but he's commander in chief. and he confirms / dissolves parliament. so when it matters, as in war time, he can . I don't appreciate the condescension in either case. It's not like his powers are easy to understand. They exist, though not commonly exercised.
I looked it up, and you are kinda right in the theory of it, and I was wrong in the practice of it. The practical power lies with the Cabinet.
From Wikipedia:
"A declaration of war by Canada [...] is an exercise of the royal prerogative on the constitutional advice of the ministers of the Crown in Cabinet and does not require the direct approval of the Parliament of Canada, though such can be sought by the government."
reply