It makes sense because virtually all machine-learning algorithms work better, and can learn faster, if you have more data.
Think of it this way: if Tesla wants to test a particular algorithm for a particular driving situation, they can "play it back" over an enormous amount of real-world situations. They will have tons more potential edge cases with which they can validate their algorithms.
Where exactly is all this data being stored and transferred? Nowhere. The car doesn't have the storage and it doesn't have the bandwidth and all indications are it isn't transferring anything that amounts to actual images of the many many cameras it has.
Sure, they can push a beta algorithm to cars and record high-level decision making between human & algo, verifying it's not totally out of whack. But that's hardly something that is going as training data into the models.
This used to be true. Modern machine learning needs way less data. The classic example is taking images and then transforming them in hundreds of ways (scaling, rotation, skew, etc) for training.
Big data is no where near as much a competitive advantage as it was three years ago. It seems not everyone outside the field has noticed that though.
I wonder if this would be true in the case of self-driving car algorithms, though (which I know nothing about). It always seemed like the hard part about self-driving cars was the 0.1% edge cases where something out of the ordinary could result in a catastrophe if not handled correctly.
Image classification seems like it would be very different, most importantly that 99.9% "correct" would be a great achievement, but for self-driving cars a .1% failure rate would be completely unacceptable.
Do you have more examples of "big data is not as much a competitive advantage", in the form of articles or research? I'm not in this field, but it's a fascinating development. It would be interesting to see to which degree it helps to perform automated transformations to increase the value of each piece of training data.
Thank you for the down-votes. I guess all the experts on HN know how easy it is to simulate training data because they all took the 101 course on how to rotate/resample images. That is uniquely a image classification technique.
Please oh wise ones how do we simulate nlp data, numeric data, finance data, biological data and anything else machine learning is used for.
Oh you are able to classify dogs and cats in images after a 2 hour youtube. How nice.
Renesd is correct that "big data" is overblown. There are diminishing marginal returns - you need orders of magnitude more data for the same incremental gain (and this blows up well beyond however millions of cars Tesla can hope to run).
You're correct that data augmentation is only a marginal technique to squeeze out more performance, and not generally possible in many domains.
You need to think this through more carefully. What would "playing it back" give you? Presumably all that would give you are a bunch of incidents where the human driver disagreed with the algorithm's output. That's it. We don't know whether the algorithm is actually wrong. Maybe the driver made a mistake. Maybe the driver wanted to make an illegal turn (very common). Maybe the driver was lazy and made a rolling stop. A human still needs to go through each incident that is flagged and manually label it.
It makes me wonder, though, if Google is currently at a disadvantage because their fleet of self driving cars is teeny compared to what Tesla has and will soon have. That is, Tesla has tens of thousands of cars with Auto Pilot sensors on the road, and (afaik) they have access to all of that real-world data to train their algorithms.
A big advantage I see in Tesla's strategy is that all of their cars are now shipping with full self-driving hardware. Even if that hardware isn't actually used to control the car, Tesla will have an order of magnitude more real-world data than anyone else.
But can they actually send all that data back to their data centers? If they are really capturing that much data, they would need to send it back using consumer home internet connections which doesn't seem realistic. I don't think the fleet size is as large an advantage if you can't actually get 100% of the sensor data.
But I think your explanation sort of highlights why promises of "safe nuclear design", at least in my opinion, have a tendency to be overstated by nuclear proponents, and then result in a complete backlash against nuclear when things go wrong.
Fukushima was a Level 7 disaster, on par with Chernobyl, that will result in large swaths of Japan being uninhabitable for lifetimes. Trying to make a distinction that it wasn't really that bad only makes your average Joe discount any assurances that "passive safe" is safe at all.
> Fukushima was a Level 7 disaster, on par with Chernobyl
This is disinformation. It's an arbitrary bureaucratic designation (Level 7 means "measures were taken to mitigate effects"), and the two events were incredibly different.
> that will result in large swaths of Japan being uninhabitable for lifetimes.
This is completely false. I've said before that if it were legal for me to do so, I'd pack up and move to the Fukushima site tomorrow. I would never say the same thing about Pripyat.
I am sure I am. But that is not the point. I gave other examples too, and the point is that requiring "real" names breaks down in all kinds of different cases, most of which by themselves are probably not that common, but that adds up to affecting millions of people.
E.g. which name for my ex is most "real"? Her legal name? Using that would prevent most of her friends from finding her - it's not the name she uses. This is common for vast numbers of immigrants with names that are hard to pronounce.
> This is the difference between a social network meant for the general populace and one specific for tech folk.
Facebook is trying to be the one to decide how we should interact with our friends, and fails to understand even how names are used in real life, not just for "tech folk".
Agreed. I'm a fairly introverted person and participate in a lot of online communities, but I rarely consider people I haven't met to be my friends. That just seems strange to me. I'd even venture to say that in the tech folk world that is a minority position. Most of the people I know need some sort of tangible validation to form a real friendship.
Pretty sure he's not. There are "personal use" exemptions that seem like they would apply. Not sure though, but if in the past ~20 years, if there is not a case of someone breaking the law by using a gopro or other camera for personal use, seems like it's a non-issue.
I'm sure they're just intending for this to be an initial product, but I agree, I don't understand their business plan at all. The people willing to pay 3x over a "good enough" backup camera are also the people who'd be first in line to buy a new car with all the bells and whistles in the first place. The market for this product has to be exceedingly small.
What would they be offering to that company, they haven't shown any prowess for integrating into the cars systems. Besides don't most car manufactures have self-driving features under development/deployed already?
Come on, that's just silly and it's a deliberate strawman. The options are not just between a $500 camera and a brand new car. The whole point is that the type of person who eschews the myriad of much cheaper camera options is also the type of person least likely to own a car that doesn't already have a built-in backup camera.
I don't understand your point about statins. The Bloomberg article you linked said statins were very effective in preventing heart attacks, just not necessarily due to their cholesterol-lowering effect. If that's the case, statins would still seem to be warranted in a lot of people.
For people who do not already have heart disease, statins have no apparent benefit. For people who have had a heart attack, statins are indeed useful. Not because of its cholesterol lowering effect, but due to its ability to dial back inflammation in arteries. For more on this here's yet another Bloomberg article on statins:
https://www.bloomberg.com/news/articles/2008-01-16/do-choles...
Buffet has said this about hedge funds for a long time now. Their fee structure incentivizes volatility, not long term performance, which is a bit ironic because the reason for the name "hedge" fund is that they should have the ability to do better risk management.
Suppose the fund gets 2% of all profits in a profitable year and gets nothing in an unprofitable year.
Scenario 1: 100 million dollar profits for two years in a row. Fund gets 4 million, customer gets 196 million.
Scenario 2: 300 million dollar profit in the first year, 300 million dollar loss in the second year. Fund gets 6 million, customer ends up with -6 million.
On a large scale, what this means is that the fund can look for large ups and downs because they gain every time there's an up, but they don't lose when there's a down.
Now if you parallelize this (so it's happening at the same time with many different customers), it means they can focus on risky bets. Maybe half their customers will win and half will lose, but the fund makes money on the winners and isn't hurt by the losers.
Most of the arguments in this comments sections are outdated. The same goes to the 'Buffet argument'. Most of the funds (after 2008) run with High-Watermark rule. Thus, in the case of your Scenario 2, fund will get nothing from the next 300 million it makes (apart from fixed fees). Thus, the investor will end up with 294 million at period 3. Just like if the fund made 100 million for 3 periods consequently.
That's an incentive of any business owner that loses money. It's the duty of investors to review the managers reputation and not give money to known "fraudsters" (unless they can negotiate an incentive structure to prevent such problems).
How so? If you're running a widget factory, you can't just close it and open again with the same machines and staff under a new name. Your investors would have good reason to sue you. And in any case you're not paid a proportion of your shareholder's returns.
With a fund, which is a separate vehicle advised by a management company, the fund can be closed with the knowledge (typically code, and who is gonna know if you take that?) kept and used to run another fund.
And it needn't be terribly blatant. If a guy has a reputation as a stock picker, he can do something like open a new fund for another geography or sector.
You don't even have to move it. You can set up a pre-pack insolvency, then buy up the business - assets, employees, stock, everything - in a new company and just keep going, but without the debt. There have been some scandals about business owners doing this, but I don't think it's been entirely effectively stopped.
Exactly. This is the real problem here, not the fairness of the fee structure. There is vanishingly little "small dollar" money in these funds. The median investor dollar in a hedge fund is a big player, and expected to be sophisticated about these things and to understand how the risks work.
The harm is to the market as a whole, not the poor investors who get hurt in a down market.
Yes there is. Volatility is an inefficiency, by definition. Efficient markets seek to the correct market price and stay there. A "volatile" market is one where parasites (c.f. hedge fund maniacs) siphon off capital due to games like this.
If we let society keep a list of facts that may not be questioned, it will inevitably have crap like "the sun orbits the earth" and "a god named zeus created the universe" that need to be questioned.
When you declare an answer to be wrong and suppress it, the only difference is whether you go on to murder/imprison/exile the heretic or merely silence them.
I think the difference between murdering someone and not putting their website at the top of a search result page is pretty significant. That the Holocaust happened is a fact. Giving correct responses to factual queries should not be a controversial goal.
When I type 5 + 6 into Google search, are people being oppressed when the top result is 11? Would this be similar to murdering ideological opponents if my page declaring the answer to be 13 wasn't at or near the top of the results?
I don't think I've ever typed in a search query before using the word "holocaust", but my results are largely the the same as the author's - Stormfront is #1, "The Holocaust Hoax" is a little further down at #5.
I think the scary thing about this query, though, is that for better or worse people DO search Google for factual answers. Sure, someone searching "Did the holocaust happen?" is likely to already be doubting the Holocaust actually occurred, but given that I think it's a very bad thing that the top answer from Google confirms their false beliefs. It turns from "Some crazy bullshit white supremacist website told me the Holocaust didn't happen" to "Google told me the Holocaust didn't happen."
If you will fact check claims about Holocaust, you will see that numbers are made up. My mother and grandmother are Jews. My grandmother lived near to Kyiv (Kiev) at times of WWWII. When I asked my mother about Babi Yar (Old Women Rave)[1], where 200 thousands of Jews were executed, she said that it is false: 190 thousands of Jews were evacuated to East, 40 thousands left only in Kyiv and surroundings, including my grandmother. I fact checked that and found that in Babi Yar only 621 Ukrainians, members of OUN, were executed by German army as partisans, including Ukrainian poetess Olena Teliga. Their grave is visible at aerial photos from USA archive made by German pilots: it near to radio mast. However, Russians claims that 200 thousands of anonymous Jews were executed in Babi Yar by Ukrainians, members of OUN, i.e. complete opposite to the fact. I also fact checked other similar claims about Odessa and Lviv and found that they are false too: in Odessa it was done by communists shortly after capture of Ukraine: thousands of victims were sink in Black Sea, in Lviv it just false: 200 dead instead of 200 000, none of them was Jew, they were British soldiers. It makes me to think that Russians are using Holocaust to meet their own goals, which makes it look real victims and real fact less trustworthy. I checked that and found answer in leaked Russian book for FSB newcomers: in Russia, state sponsors fake campaigns to help FSB to develop new agents in various organizations. Book describes technique how to use these fake claims to find a someone who is easy to manipulate. I was shocked by that. So, I know that Holocaust is real thing, because some of my grandmother relatives in Poland were victims of that, but I also know that some numbers are made up or fake. How I should react to all that?
Please, if you downvote, downvote with comment. I can provide documents, photos, records if necessary. It's real story of real people. I'm sorry if you are victim of Russian propaganda. I also not pretending that 621 members of OUN were the only victims. I'm talking about Babi Yar only. Thousands were executed at various places across and near to Kyiv, some of them were Jews, but none of Jews were executed in Babi Yar, AFAIK. I live here so I know better.
PS.
It was hard time for Jews. My grandmother was Jew and her husband was in Soviet Army, so she was valuable target, however she was hided by neighbors, so she escaped. Some of her relatives was not so lucky (I know no details, except that one of them ended dead in Poland), but none of them were executed in Babi Yar.
Think of it this way: if Tesla wants to test a particular algorithm for a particular driving situation, they can "play it back" over an enormous amount of real-world situations. They will have tons more potential edge cases with which they can validate their algorithms.