SHITPOSTING_ACCOUNT @ SHITPOSTING_ACCOUNT @feddit.de

Posts

2

Comments

193

Joined

2 yr. ago

How much software is still running 32 bit binaries that won't be recompiled because the source code has been lost together with the build instructions, the compiler, and the guy who knew how it worked?

How much software is using int32 instead of time_t, then casting/converting in various creative ways?

How many protocols, serialization formats and structs have 32 bit fields?

Signed integers. The number indeed goes to 2^32 but the second half is reserved for negative numbers.

With 8 bit numbers for simplicity:

0 means 0.
127 means 127 (last number before 2^(7)).
128 means -128.
255 means -1.

The most common date format used internally is "seconds since January 1st, 1970".

In early 2038, the number of seconds will reach 2^31 which is the biggest number that fits in a certain (also very common) data type. Numbers bigger than that will be interpreted as negative, so instead of January 2038 it will be in December 1901 or so.

The server uses a kilowatt of power or more (most of it in the CPU). But if the server is serving 1000 active users concurrently, and only 5% of the time you spend online is spent fetching ads, 20000 people staring at their screens get their ads from let's say 2 kW of server power usage, plus another 2 kW for all the equipment to get the data there... for a total of 0.4 watts per user.

These are completely eyeballed numbers, and could easily be off by an order of magnitude.

But your on premise gear (screen, computer, router) are likely by far the biggest factor.

One easy way to cross-check power usage claims is cost. It will only catch the most egregious bullshit, but it's easy. A random page I found claims that "According to the American Council for an Energy-Efficient Economy it takes 5.12 kWh of electricity per gigabyte of transferred data."

A Steam game with 50 GB would thus consume 256 kWh. Even if your 300 watt idle gaming rig, 50 Watt Router and 150 watt screen to watch the progress bar spends 2 hours downloading that, that's 1 kWh. Even at 8 cents per kWh, that means just downloading the game would cost someone (not you) over $20. Do you think steam would let you delete and redownload that game that you bought on sale for $10 as much as you want if between them and your ISP someone had to pay for $20 just in electricity, each time? Not the game rights, not the servers, not the connection, just power.

That's one possibility. It's also possible that you have decent privacy settings keeping them from knowing too much about you, or they simply use a shitty ad network that's bad at targeting. Even the major ones are impressively bad.

There also aren't many advertisers interested in these ad slots since they know people watch them only for the reward, and games are also a frequent source of ad fraud (I think), so serious advertisers avoid them.

Also, mobile gamers are likely not the most attractive audience for the high paying stuff.

There are stock news site that churn out "why did $STOCK move in $DIRECTION" filled with bullshit speculation. I bet it was mostly automated even before chatGPT and has gotten much worse now.

The ad categories offered by various companies vary and I think adsense is nowhere near the closest-targetable network there is.

Try showing an ad to only Python software developers. Not IT repair shops. Not software developers writing exclusively C. I think you may be able to do that with keyword targeting on AdWords, once you avoid the bear traps you mentioned, but it's hard.

OTOH, I bet there are ad companies that will help you target "30-40 years, single, lonely" for dating ads (that might be possible even with adsense), and definitely people with specific diseases to peddle medicine to them.

Occasionally someone posts a list of categories used by one or multiple networks and they can be the most specific, or far too broad (see: python dev).

I'm extremely surprised that I haven't seen ad companies offering specifically to advertise to people working at specific companies. I'm sure it exists, just haven't seen it. This would be incredibly valuable both for job ads, industry specific ads (this would benefit from breaking down by department), and also criminals and spies trying to get people from specific companies infected with malware.

What's also important to understand is that these categories don't need to be accurate. "This person has a 80% chance to be in category X" is more than good enough. Hell, 10% would probably already work.

The right ads pay really well. A life insurance click can be worth tens of dollars, because the conversion is worth thousands. So if there is a 10% chance you're interested in buying life insurance, bombarding you with those ads makes sense.

Those are different models. Ads can be sold pay per view, pay per click, or even pay per conversion (the store reports when the customer buys something and only pays for that).

These can be converted by multiplying with the estimated probabilities. For example, if the scammer is willing to pay $1 for the click, and the probability that the user will click is estimated to be one-in-500, the view would be worth 0.2 cents.

If the scammer is willing to pay $20 for the conversion (because it means they successfully scammed someone out of $30), they'd need to succeed scamming one in 20 users that clicked for this to work out.

Works the same for legit businesses of course, where the business will consider total lifetime value (not just the current sale - you might also subscribe to something and keep paying for 2 years, or come back to buy again). Advertising / customer acquisition costs are a huge part of many businesses, which is why running online ad platforms is so obscenely profitable.

In this case, I don't know who in the chain will do the conversion - if the bid will be for a click and the ad platform will estimate how likely you are to click, or if the bidder makes the guess and bids based on that. The bidder in this case would be another ad platform of course, acting on behalf of the actual advertiser, and nobody in this "ecosystem" trusts each other. It's full of companies trying to scam each other or companies offering services to validate that the data someone is feeding you is real.

I get the joke, but for those seriously wondering:

The epoch is Jan 1, 1970. Time uses a signed integer, so you can express up to 2^{31 seconds with 32 bits or 2}63 with 64 bits.

A normal year has exactly 31536000 seconds (even if it is a leap second year, as those are ignored for Unix time). 97 out of 400 years are leap years, adding an average of 0.2425 days or 20952 seconds per year, for an average of 31556952 seconds.

That gives slightly over 68 years for 32 bit time, putting us at 1970+68 = 2038. For 64 bit time, it's 292,277,024,627 years. However, some 64 bit time formats use milliseconds, microseconds, 100 nanosecond units, or nanoseconds, giving us "only" about 292 million years, 292,277 years, 29,228 years, or 292 years. Assuming they use the same epoch, nano-time 64 bit time values will become a problem some time in 2262. Even if they use 1900, an end date in 2192 makes them a bad retirement plan for anyone currently alive.

Most importantly though, these representations are reasonably rare, so I'd expect this to be a much smaller issue, even if we haven't managed to replace ourselves by AI by then.

My guess is that it's a couple watts while you're actively using the internet, mostly due to the extra CPU load a few bad ads cause when they're on your screen. Without having done the math I expect all the servers, data transfer etc. to be negligible, on a per-user basis, because they serve so many users.

That's another interesting thing btw. Most of the "internet thing X uses Y amount of electricity" are utter bullshit and massively exaggerating. What uses most power on desktop/TV is the screen. The second biggest consumer is likely your router (which is on whether you use it or not, but the studies usually ascribe all of the standby usage to your active usage - this makes sense if you try to look at "how much CO2 does all our digital stuff including 'having an Internet connection' cause" but not if you're trying to look at "how much extra CO2 does activity X cause, assuming I already have an internet connection because I'm not gonna live in a cave").

How online ads actually work.

Very simplified TLDR: you visit a news site. They load an ad network and tell it "put ads here, here and here".

The ad network now tells 300 companies (seriously, look at the details of some cookie consent dialogs) that you visited that news site so they can bid for the right to shove an ad in your face.

One of them goes "I know this guy, they're an easy mark for scams according to my tracking, I'll pay you 0.3 cents to shove this ad in their face". Someone else yells "I know this guy, he looked at toasters last week, I want to pay 0.2 cents to show him toaster ads just in case he hasn't bought one yet."

The others bid less, so that scam ad gets shoved in your face.

That's extremely simplified of course. https://en.wikipedia.org/wiki/Real-time_bidding has a bit more of an explanation.

Don't do stupid shit because the Internet tells you it's a challenge.

The next time it may not be a chip but a tide pod. Or "crystals" made by blowing bubbles with a straw into a bucket of bleach and vinegar (the blowing makes sure that the victim takes a deep breath of the World War 1 gas warfare recreation they just mixed up).

This is VERY country specific. In some countries ambulances focus on fast transport with minimal care in the ambulance (IIRC this is the case in the US), elsewhere they can provide significant first aid while on the way. If it takes you 15 minutes to the hospital and the ambulance needs 10 to get to you and 10 to the hospital, you'll be at the hospital 5 minutes later but will receive care 5 minutes sooner.

In Germany the ambulance will have what I think would be equivalent to one EMT-B and one paramedic, but a emergency physician may be brought to the scene with a separate car.

Because I've literally been told that by hospital staff (in Germany). This may vary by country.

I've heard that no matter how often you tell a kid the stove is hot and will burn them, they won't stop trying to touch it until the pain has taught them. Not sure if it's true (or true for all kids), but I would expect the other side of that ("once they've burned themselves, they learn") to be mostly reliable.

Tell that to the custom binary serialization formats that all the applications are using.

Edit: and the long-calcified protocols that embed it.

For the marketing folks: adding "AI" to your product ad may increase the chance that the pencil pushers will want it, so I get why it may make sense to put it there. But it will make IT folks start with the assumption that your entire product is worthless bullshit that tries to trick people into buying it with meaningless buzzwords. Same for "Blockchain".

If I had a product that actually had a good use for AI/ML, I'd use the most technical term possible to describe it, just to avoid the appearance of buzzword fishing. With blockchain... just invent some new name for it. It's so toxic that people will roll their eyes and stop listening.

With most pre-written SLAs, the penalty is something like "we'll refund the service cost for the month" at best. So it's "we have a financial reason not to fuck up" not "you will be made whole if we fuck up and your business is down".

The SLAs are also often tied to SLOs (the quality they promise to deliver, e.g. "we promise to be up 99.5% of the time") that are very generous for the service provider. If your critical service was down 3.6 hours in a month, that would still meet a 99.5% SLO. So if your business was down for 2-3 hours per month, that would be a-ok. Only if it was down for say an entire day, you'd get (depending on the contract) typically either a day or a month of service refunded.

I'd take a provider with no SLO but a good track record over someone that offers an SLA. If they fuck up the month of refund is going to be the least of my problems, and if they fuck up repeatedly, I'll have to emergency-migrate away to a different provider either way.

Y2038 is my "retirement plan".

(Y2K, i.e. the "year 2000 problem", affected two digit date formats. Nothing bad happened, but consensus nowadays is that that wasn't because the issue was overblown, it's because the issue was recognized and seriously addressed. Lots of already retired or soon retiring programmers came back to fix stuff in ancient software and made bank. In 2038, another very common date format will break. I'd say it's much more common than 2 digit dates, but 2 digit dates may have been more common in 1985. It's going to require a massive remediation effort and I hope AI-assisted static analysis will be viable enough to help us by then.)