joefkelley

I happen to have worked on exactly this at Google. No, we don't train on restrictively-licensed code to the best of our abilities.

This reminded me of one of my high school computer science assignments- simply to find all words in a single boggle board. And try to optimize your solution a bit. The point was to teach about recursion/backtracking and data structures. The intended solution was roughly: start at a square, check if your current prefix is a valid prefix, move to a neighbor recursively, and emit any words you find. Trying to optimize naturally motivates a trie data structure.

I found it to be at least an order of magnitude faster, though, to invert the solution: loop through each word in the dictionary and check whether it exists in the grid! The dictionary is small compared to the number of grid paths, and checking whether a word exists in the grid is very very fast, requiring not much backtracking, and lends itself well to heuristic filtering.

Yeah I guess wokeness and cancel culture are what you complain about now when your life is so free from challenges that you have nothing else to complain about.

The burden of proof is on the person making a positive claim. So I didn't think I really needed any more argument than "the article's argument is insufficient for me to believe their claims". I wasn't trying to make any more general point about censorship or social media.

I don't have a problem in general with a ticketing system for the government to request specific kinds of moderation. There do exist valid carve-outs of free speech, and I see no reason why government and industry couldn't collaborate on that. If it were used to censor political speech, then yes, I would have a problem with that. But I haven't seen evidence of that.

The evidence presented in the article does not match the claims.

The emails they include show that there were meetings between DHS and Twitter and between DHS and Stanford, on the topic of election integrity. And that there was a Signal chat (I guess this is kind of sketchy).

But there's no evidence of censorship or anything politically-motivated that I can discern.

Do you feel that a language has to have something at the language level to prevent NPEs?

In my experience, Scala does pretty well without it.

I guess is your point that the language should make it impossible to write bad code, not just make it easy to write good code?

3) Don't transparently download and install stuff without user interaction, regardless of where it comes from!

This is an interesting one. I totally get your point. But also users are terrible about updating their software if you give them the choice. Automatic updates have very practical security benefits. I've witnessed non-technical folks hit that "remind me later" button for years.

If you buy youtube premium you'll very likely see more ads for youtube premium

Really? I have youtube premium and I can't recall seeing ads for it. Why would they advertise a product to people that already have that product?

Google Ads could have a flag that turns it's data collection/sharing off and replaces it with micropayments to any site that user visits that are running Google Ads.

FWIW, this flag already exists, except you don't have to do micropayments: https://adssettings.google.com/

I guess you still see ads with this setting, they're just not personalized. Hypothetically you could imagine a "stronger" setting that doesn't just do away with personalization, it does away with ads altogether by allowing the user to "outbid" any advertiser. But I suspect there will be some surprised users who get a bill for hundreds of dollars by doing some particularly high-value searches like "personal injury lawyer" or "mortgage" or something.

And if it were a flat rate, my intuition is that the fee would have to be much higher than most would expect or be willing to pay.

I'll hedge this by saying I can only speak for the teams I've worked on, plus my somewhat limited understanding of company-wide policies.

The answer to both questions is no. If you opt out, your data is not used for modeling or targeting or anything. Perhaps some internal reporting that isn't used for anything other than like PMs wanting to understand user behavior? Even that I'm not sure about.

If you are identified as being in market for something based on activity and then opt out, you will no longer be classified as being in market. That classification will be deleted- though perhaps not immediately but within some reasonable time frame, say 24 hours or so.

I'm a Google engineer who interacts a little bit with this kind of stuff.

There are a few settings for this kind of thing. You can take a look at https://myaccount.google.com/data-and-personalization

The "Ad personalization" bit is probably what the parent comment is referring to. But it sounds like you're interested in the "Web & App Activity" bit, which will turn off the non-ads usage of your data. To a certain extent at least, since there are some grey areas.

For example, I'm on the team that sends Google Shopping emails. If you click a button to track the price of a specific TV, we'll still send you an email if that price drops even if you've opted out of "Web & App Activity". But if you've just been browsing shopping pages for TVs, we won't send you a general email about TV deals if you've opted out. Both of those cases are in some literal sense "web activity" but it's still pretty clear what the user expects.

But you might imagine- if you're tracking the price of a specific TV and opted-out of Web & App Activity, should we send you an email if a near-identical TV drops in price? We probably wouldn't, and we don't have anything like this today, but it's not quite as clear. And Google has so many features across different teams, I can imagine there's probably at least one where some privacy reviewer made a different call than what you would have made.

I think each couple has a slightly different balance on what level of collaborative decision making they can expect, and this is actually a big factor in compatibility.

For instance, I take your approach for most purchases under two thousand dollars or so. If I want to buy myself a new computer or whatever, I'd mention it to her, but ultimately I'm probably going to get it even if she thinks I shouldn't. I know some couples where this isn't the case, even if they have the means. Their price threshold for making the decision together is much lower.

But on career changes we make decisions together. For instance, she recently made a change that will result in her making less money, especially in terms of long-term career trajectory. But her stress level and overall happiness is much better. And she knows that my income was a good amount higher anyway and it ultimately won't affect things like when we retire or our quality of life that much.

But then it would be pretty shitty of me to change to a lower-paying profession down the road without her OK. She has sacrificed her earning potential with this kind of commitment in mind and maybe wouldn't have if she didn't know she could count on me to make future decisions with her collaboration.

I'm not saying either end of the spectrum is necessarily better. Just that there are pros and cons and it's more important to be in agreement.

Ah! Thanks for explaining, that makes sense.

I could be wrong (I'm not a finance guy) but this isn't my understanding of how fractional reserve banking works.

If Apple deposits $200 billion and the ratio is 1/10, then the bank can loan out $180 billion of that 200; it must keep $20 billion in reserve - 1/10th.

In your scenario, the bank has negative $1.8 trillion in reserve.

I'm with you 100%. I thought censorship meant "not allowing you to speak". I guess now it means "not actively helping you profit from your speech".

1. I am not sure. This is all just my layman's interpretation. I'll admit I could be very wrong and have no expertise. I'm just discussing. 2. No, in general I believe there's a difficult balance to be struck between individual rights and collective interests. I tend to lean more toward individual rights actually.

I would love it if local governments would stop restricting the rights of local developers and allow the free market to determine what is built where more often.

Roughly, we have three types of color-sensitive cone cells in our eyes. They each have different behavior in terms of how much they "react" to different wavelengths of light. Take a look at this chart: https://en.wikipedia.org/wiki/Trichromacy#/media/File:Cones_...

Each individual wavelength activates all three to some extent: think about this as a point in 3d space. For example, 400nm corresponds to something like (0.1, 0.05, 0.0), from that chart. 500nm might be (0.1, 0.4, 0.3).

But we experience a mix of many different wavelengths at once. So we can not only experience just these points in 3d space, but we can also experience any linear combination of any of these points. For instance, a mix of half 400nm and half 500nm light might be "sensed" by us as (0.1, 0.225, 0.15), even though maybe there's no individual wavelength that corresponds to that point. Any linear mix of any number of any wavelengths covers the entire gamut of what we can perceive.

The question then for someone picking primary colors for an additive display is: if I can only do a linear mix of three wavelengths, what wavelengths should I pick? What covers the biggest subset of the whole perceptible gamut? It just so happens that red, green, and blue do the best.

If you swapped green for yellow, there would be a section in that 3d space that you could no longer create. Specifically, the area where M cones are strongly activated compared to L and S. Unsurprisingly, this would be the greenest greens.

The tough thing is of course you are right. It's obviously in your best interest to vote against new housing if you own a house. But it's not in society's best interest.

I think some of this might be misinformed.

I know Hive can definitely query other datasources like traditional SQL databases, redis, cassandra, hbase, elasticsearch, etc, etc. I thought Impala had some bit of support for this as well, though I'm less familiar with it.

And SparkSQL can be run on a multi-tenant cluster with automatic resource allocation - Mesos, YARN, or Kubernetes.

It's been about 4 years since I've been in this world, but I remember there being several products all doing a very similar thing: Presto, Hive, SparkSQL, Impala, perhaps some more I'm forgetting. Is the situation still the same? Or has Presto "won out" in any sense?

People tend to focus on what has been left out, but think about what they actually did learn about:

Drafting from that pool, item and skill builds, last-hitting, creep aggro, laning in general, jungling, item and spell usage, ganking, team-fight positioning, pushing objectives, warding, map control, farm priority, when to retreat vs engage. All of these require an understanding of micro vs macro goals and how they relate.

Surely this qualifies as "a difficult task."

"keming" is the only one-word joke I know of.

Wondering if anyone is aware of any others.

I worked at one (horrible) place where we explicitly only deployed on Friday afternoon.

The logic was we have fewer users over the weekend, so if something goes wrong, fewer people will notice.

Nope, doesn't work at all for me either.

I read the article, thought to myself, "let's see how HN finds a way to say this is actually bad for privacy", clicked through to comments here, and was not disappointed. The hivemind anti-Google kneejerking is quite out of control.

I'm an engineer who has worked on ad systems like this and I'm really struggling to make sense of this article - what hope does a layman have?

Here's my understanding: Google runs real-time bidding ad auctions by sending anonymized profiles to marketers, who bid on those impressions. The anonymous id used in each auction was the same for each bidder, which is in violation of GDPR. If Google were to send different ids for each bidder, it would be ok? Is this correct?

Why would it matter that the bidders are able to match up the IDs with each other, aren't they all receiving the same profile anyway? Wouldn't privacy advocates consider the sending of the profiles at all an issue?

I understood the title slightly differently: the ultimate conclusion was a general pattern of conscientious health behavior is the real predictor. So those that do at least one push-up are likely to be healthy, even though that is purely correlation and not causation. All of these other metrics that have surprisingly high correlation are indirectly measuring whether you generally live healthily.

I'll try to explain why I'm excited: I think you hit it with the "anyone in the world" desire. If I truly want to communicate with anyone I know, I have three options: call, text, or email. No other messaging platform has anywhere near 100% coverage for the people I talk to. RCS won't initially have that coverage, but I do see a path to reaching that coverage since it doesn't require every single user installing and configuring an app. I'm already pretty happy with SMS, I just want some more features.

If lack of end-to-end encryption is the price to pay to get carriers to implement the features I want - better group messages, read receipts, better quality photo sharing, typing indicators, etc - then sign me up.

And Google taking the lead by implementing this for free seems like it should put a stop to any worry that carriers will charge for it.

I think bidding tends to just be about what the advertiser cares most about. If they are a "brand" advertiser (Coca-Cola is the canonical example) they might bid per impression since they're not really looking for clicks, and they have some idea of how much an impression is worth to them. Compared to say some app where they want installs and so might bid per conversion.

As for the network, it's also not as clear what's profitable. Bidding per click or conversion probably gives them more opportunity to do well with good targeting. But high per-impression bids are also useful for the users the network knows very little about.

Advertisers go to an ad network and configure their campaign with: "I want to show X to users who are Y or are on a site about Y, and I'm willing to pay Z for it".

X is just the content of the ad and is usually entirely advertiser-defined. Some ad networks will allow you to do some fancy stuff to help determine the best phrasing or arrangement or whatever of your ad.

Z can be in dollars per impression, per click, or per conversion. An impression is one person being shown an ad one time. A click is a person clicking an ad. And a conversion is advertiser-defined but is usually someone buying the product being advertised. Bidding per click is the most common.

Then when a user loads a page with an ad on it, the ad network finds all campaigns that are eligible - meeting the criteria defined in Y. It then runs an "auction". It has to calculate which eligible advertiser is willing to pay the most for that potential impression. But not all advertisers are bidding per impression, so some prediction comes into play. If an advertiser bid per click, then the expected amount they will pay is the probability the user will click multiplied by the price if they do click. Their probability of clicking is often called CTR which stands for "click-through-rate" so that's where the bid * CTR comes from. Usually an ML model takes what is known about the user, the ad, and the page, and predicts CTR. Similarly for conversion-based bids, it's bid * CTR * CVR, where CVR is "conversion rate". Whichever ad comes out with the highest result of this calculation is shown.