joefkelley

The idea that ad targeting sucks because ad networks have no incentive to improve it is extremely dubious.

Google and Facebook have an incentive to use up advertisers' budgets. But they also have an incentive to keep advertisers happy by providing good returns so that they will keep spending / increase spending in the long run. Advertisers are not always perfectly rational, but I don't think they're as dumb as the article makes them out to be.

The simpler explanation for poor ad targeting is the bidding component. If it was the case that you were shown whichever ad had the highest predicted CTR, I'm pretty sure you'd be seeing some damn relevant ads. But instead you're seeing whichever ad has the highest (bid per click * ctr). Or worse, advertisers can also bid per impression, in which case there might be no relevancy component at all.

My guess is the car ads on YouTube mentioned in the article are that last case. Both of those components (YouTube and car companies) are notable for skewing towards "brand" advertising rather than "performance" advertising. They're not trying to get you to buy a car right then. They're trying to get you to feel good about their company in the long run so that five years from now you're more likely to buy their car.

I think I'm not understanding if you really answered the question you quoted - what specifically is the concern with having that information extracted? Is it just "it feels creepy"?

My point is that they already have the information, so any potential misuse or abuse is possible regardless of what types of parsing they're doing. It might just be inconvenient.

I'm usually much more pro-google than most on HN, but this is quite bad. I understand it enables other features and a technical work-around that also keeps those features is difficult. But this is a disastrous user experience.

It reminds me of caller-id spoofing. Yes, there's a legitimate use case for it. But don't just throw security out the window to enable it, especially when there is a clear and obvious way to abuse it.

Wait, so you trust them to store the email, and parse it for something like spam detection, but not for some smarter feature? What specifically are you concerned about that is a problem with this and not with them having the email in the first place? Sounds like you shouldn't be using Gmail at all tbh.

Oh everyone is aware of it. They have been making some changes to the promotion process to try to address it. But it's still a huge issue.

I'm not so sure about what the alternative is, though. The promotion process at other places I've worked was pretty much "does your boss like you? If so, they'll promote you." Which has a whole host of other issues. For instance you can get promoted for doing a mediocre job at maintenance work but being very friendly and likable.

Context:

In ~July 2018 there was some outcry because Google was "letting third parties read your emails" (e.g. https://www.cbsnews.com/news/google-reportedly-allows-third-...). Of course, these were all explicitly installed by users who gave these apps access. But somehow people were mad anyway - maybe users shouldn't be given the option to make choices they don't understand?

Anyway, as the message mentions, Google announced new requirements for these apps on October 8: https://cloud.google.com/blog/products/g-suite/elevating-use...

Apparently, IFTTT (which does personal automation, integrating with many third parties), does not comply with the new policy.

Sorry, maybe I missed it; where in the announcement did they say it was a subscription?

I agree that this process sucks. Do you think it's possible to do 1 and 2 without 3? How could a business that remains focused on benefiting free users compete/profit?

When Google takes the approach you want, they are vilified for "letting third parties access your data" - as in the Gmail "scandal" a few months ago.

If you let users do whatever they want, they will often shoot themselves in the foot. It's a tough balance.

I can only speak for Stanford when I was there, but almost every CS graduate took "Computers, Ethics and Public Policy" as it was the best way to satisfy the degree's "Technology in Society" requirement.

In talking to friends and coworkers it sounds like taking an ethics course was more common than not.

Definitely agree with this.

I've been skiing my entire life (age 3 to currently 28) and a spent a great deal of time with an instructor.

My girlfriend has also been skiing her entire life, pretty much never with an instructor. Her first time she joined me and my family with the instructor, she was blown away by how much the instructor was able to improve her skiing in ~20 minutes of teaching.

Seriously, good ones know their shit.

Would you consider an ice hockey player's skill in 3-on-3 overtime or shootouts to be a factor in their overall ice hockey skill?

No idea. I guess technically I omitted:

  (2,1), (1,2): 36.86989764584401
  (3,1), (1,3): 53.13010235415599

So I don't think it's a bug in my code, but that is pretty odd.

The (7,2), (1,4) solution is surprisingly good. I did a really dumb brute force search and you have to get a decent ways out before you get anything better. Here are pairs of points for which there are no other better pairs with a lower max coordinate:

  (4,1), (1,4): 61.92751306414705
  (5,1), (1,3): 60.25511870305778
  (6,2), (1,5): 60.255118703057775
  (7,2), (1,4): 60.018360631150664
  (17,3), (4,11): 60.00891367665869
  (18,7), (2,13): 60.00333223031155
  (19,5), (3,11): 60.001318460471495
  (26,7), (4,15): 60.00009466237024
  (59,21), (9,49): 60.000006796458415
  (97,26), (15,56): 60.00000048796379

> Facebook offers a service that people wants. A service that is not morally bad, if any connecting people is a positive thing.

People also want heroin. That doesn't mean it's actually a positive thing in people's lives. If anything the evidence is mounting that it is negative, e.g.: https://hbr.org/2017/04/a-new-more-rigorous-study-confirms-t...

I mostly agree with the rest of your post, though. I believe it's possible, though very rare, to work at Facebook and have a positive impact.

And there's definitely a total mismatch between regulation and the realities of big tech companies. I'm skeptical of government regulation solving privacy issues specifically (I think that stems more from a widespread cultural misunderstanding or unawareness of privacy concerns) but maybe it's needed for other ways these companies are negatively impacting the world.

Seems like it shouldn't be too hard to make a Rule 110 cellular automaton (https://en.wikipedia.org/wiki/Rule_110), which is turing complete.

Honestly, that's fair. In my experience Google does a lot toward being trustworthy since it's in their long-term best interest. But at the end of the day I'm just some random guy on HN comments, so you're right to be skeptical.

No.

I mean, maybe high-level aggregate data like "X% of users got emails about shopping" might still be in some spreadsheet somewhere, or in people's brains. But no, I can't think of a way any personalized data would still be around.

In the past. We (I worked on gmail ads during this time) stopped doing it about a year ago.

https://blog.google/products/gmail/g-suite-gains-traction-in...

Are you talking about orange extract? or cayenne pepper? The post is here: https://ai.google/research/pubs/pub46507 and direct link to paper is here: https://storage.googleapis.com/pub-tools-public-publication-...

And the recipe I found internally is "The Real California Cookie" in that paper:

Bake at 163C (325◦ F) until brown:

• 167 grams of all-purpose flour.

• 245 grams of milk chocolate chips.

• 0.60 tsp. baking soda.

• 0.50 tsp. salt.

• 0.125 tsp. cayenne pepper.

• 127 grams of sugar (31% medium brown, 69% white).

• 25.7 grams of egg.

• 81.3 grams of butter.

• 0.12 tsp. orange extract.

• 0.75 tsp. vanilla extract.

I wonder if there is some kind of "steady state" vs "transition" effect that could explain this though. For instance, if you put up a sign that changed the speed limit from 70 to 80 but changed nothing else about the road, would you see an initial increase and then decrease after a few miles? Although I'd also guess that in reality if the speed limit changes to 80, there's probably other characteristics of the road that change so that the steady state might not regress back in this instance.

This is indeed surprising! For those like me who need a bit more detail before this clicks:

If you are going X miles, the difference in gas usage between A mpg and B mpg is (X/A)-(X/B) = X(B-A)/(AB). So in all these pairs, the constant is (B-A)/(AB), in this case 1/60. So each adjustment means your fuel usage goes down by 1/60 of a gallon per mile.

Don't forget safelane pos 1 feeding at minute 6 and typing "gg mid no gank"

I don't think it's that complicated to adapt to... everyone should just be pretty much constantly ferrying out regen and harass more aggressively.

> I've seen lots of people suggest that OOP+FP fusion is what allows Scala to provide an easy gradual on-ramp to FP for people and projects transitioning from Java, which is Scala’s key value proposition.

I feel like I'm at the tail end of that on-ramp, and while I'm still as a whole really appreciating scala, I'm starting to wonder what the end-game is.

Now that I understand and am comfortable with all the functional stuff, I'm left wondering why I still have the OOP stuff, and why some of the FP stuff isn't more natural. I guess you still need some OOP for java interop, but the pure-scala ecosystem is pretty significant already.

I guess what I'm saying is that the easy transition from Java is only a benefit for bringing people in, not for a language as a mature ecosystem. And I'm thinking maybe I should get into one of the full-FP languages.

In fact, if anything, what's keeping me from making the switch is not any OOP features, but actually just imperative! Especially for performance-sensitive sections, and some rare-but-necessary data structures, it seems you really do want a small amount of mutable state.

One way this could be calculated is to assign transition probabilities to the edges, and calculate the expected number of steps a random walk takes to reach the end. (wikipedia has the math required: https://en.wikipedia.org/wiki/Absorbing_Markov_chain#Expecte...)

I'm not sure how best to assign probabilities, or how sensitive it would be to that. Perhaps uniformly, or perhaps biased toward moving closer to the end state.

Quill is really neat and I agree that it's a lot further down the road than most libraries like this.

But it all feels a bit too "magical" for me. Seems like you're trading one class of bugs - hand-written SQL mismatching your types - for another - this magic doing something unexpected when you try to do something weird. But maybe it's really really good and that second class of bug never happens. I'm just hesitant (and happy with doobie for now).

Seems it's five individual agents with no communication, just a reward function that shifts towards team-based rewards:

"OpenAI Five does not contain an explicit communication channel between the heroes’ neural networks. Teamwork is controlled by a hyperparameter we dubbed “team spirit”. Team spirit ranges from 0 to 1, putting a weight on how much each of OpenAI Five’s heroes should care about its individual reward function versus the average of the team’s reward functions. We anneal its value from 0 to 1 over training."

So pretty much like pubs.

You might be able to trace the execution of the iterative code more easily, but in my experience it is often much less clear why that produces the correct result and how that code was written in the first place.

Take a look at the wikipedia page for computing Levenshtein distance: https://en.wikipedia.org/wiki/Levenshtein_distance#Computing...

The recursive version needs barely any explanation. But ask me to carry it out by hand and I'm sure I'll pretty quickly get lost. The iterative version needs a lot more explanation for why it is the way it is, but I also think I could carry it out on paper quite easily.

It's not clear to me how those two topics are related.

If it's about whether Google "knowingly" enabled sex trafficking, then it seems like a pretty simple argument is "sure, we knew about porn, but we didn't know about the sex trafficking" is a pretty reasonable defense.

Also, I don't know, I just work here.