• 1 Post
  • 26 Comments
Joined 2 years ago
cake
Cake day: June 9th, 2023

help-circle










  • As someone who works in advertising, that is partially true, but also not the complete story…

    Data brokers want you to believe that the more data you have the more likely your ads are to be successful, but in reality it’s not about the amount of data but the quality of the data. If you have someone who has looked at reviews of gym shoes/different models on different stores, then that data is pretty valuable as you can focus on getting them to buy from your store or try and advertise models at the top of their budget, which will likely lead to a higher ROI than just advertising on fitness forums (note it is super hard to get the balance between tipping people over the line to buy and advertising them something they were already going to buy/had already decided against - Google particularly are absolutely terrible at this, but also do evaluation in house, so they’ll misrepresent to advertisers that your ad which showed up one link above your non-sponsored link made 100% of the difference in getting the purchase). Similarly, if you have data that someone is active on a car audio forum and recently bought a specific model of car, you can advertise kits/speakers specifically to that car, which is better than just advertising “hey, we make audio upgrade kits for [specific car/cars in general] on a forum/related site”.

    This also makes advertising one of the few situations where using ML actually makes sense - there’s huge amounts of data (way more than a person can consider) to come in, and patterns which lead to good results (someone purchasing something) or bad results (someone not purchasing something). It’s not worth a human targeting every single microcategory, but if an ML model can pick up that advertising to (eg) people who have recently purchased cameras who are interested in triathlons and often visit areas with with high rainfall makes them more likely to buy your specific aftermarket lens hood, then it makes buying the ads so much more worth it and also lets you extrapolate onto other microcategories which may also have similar results, and if they don’t then that updates the model.

    Generally data is less useful for awareness campaigns (ie “next time you’re in the supermarket/in the business for x, buy our brand” type of campaign), especially if it’s already on a relevant site, but it’s still somewhat useful if someone is reading on a (trustworthy) news site or watching an ad-supported streaming service, however purchase data & activity data is still useful for showing more relevant ads, as while 90%+ of people on a fitness forum are going to be into fitness, I don’t think 90%+ of general site visitors or tv show viewers are going to be into anything specific enough to make it worth it to advertise it.




  • I made a mistake from remembering something I read a long time ago and corrected my comment accordingly. I also made it clear that I thought the sentence for the protesters was way too harsh, I just took issue at the words “peaceful protest” when people were injured as a fairly direct result of their actions.

    I definitely support JSO over the oil companies and whoever they pay off in government, but I don’t think claiming a protest is “peaceful” should be a “get out of jail free card”. Call it a justified protest or whatever, but the vast majority of what protesters call “peaceful protest” is actually not peaceful at all, as the actual peaceful protesters aren’t having to defend their agressive or unpeaceful actions.


  • I just looked up the car crashes and it was 3 people to hospital as a direct result of the traffic, I got casualties mixed up with deaths. I’ll edit my comment to say.

    There’s no exact figure on ambulances delayed, but given all the traffic it caused, but for heart attacks (all I could find good data on) the chances of surviving are halved in a 4 minute period (https://www.ahajournals.org/doi/10.1161/JAHA.120.017048), so factoring in all serious injuries there’s a decent chance that at least one outcome was severely affected if not changed to death as a result of an ambulance being delayed.



  • Even if you’re not banned from hexbear can you really interact with it?

    I wouldn’t call someone going through the wikipedia article for informal fallacies like it’s a checklist then brigading all your past activities an interaction so much as an experience, and not a good one at that.


  • it’s also worth mentioning that JPEG was designed for photographs, where there’s a very high likelihood that each pixel of an image will be a different colour to those immediately next to it. Because of this, JPEG not only has higher file sizes for text and 2d graphics/pixel art than PNG and especially a compressed SVG (which would far and away be the best method of representing the mario image - it’s close to the same “tile” thing but transferrable and well supported), but it also results in artifacts and lower quality of the image.



  • I don’t think you understand how it works… An upload:download ratio must average (not simple mean, but that’s because ratios are nonlinear - I can’t recall the mean type but it’s the nth root of multiplying them all together) 1 in a system where all uploads and downloads are logged in the same tracker. It doesn’t matter who the uploader or downloader is or how recently they made their account. That’s what I meant by a closed system.

    An open system would be where you download parts or all of a given torrent via another tracker, and the same with upload. The private tracker only logs what you downloaded and uploaded though it, so your ratio from the perspective of that tracker is different to in reality.

    Even if you ignore the first 5 files or 15GB or whatever for new users, if you have those files then great but do you really want to turn it into a betting game of seeding supply and leeching demand?