1,956 karma · joined July 24, 2024
In fact, being able to produce unlimited numbers and variations and combinations of adverts and have them compete against each other in the real world and be scored on tiny deltas in metrics becomes much more possible if you can automate basically the whole process. But it's a multimillion spend if you have to recruit actual actors and actual film crews and actual food photographers or drone pilots and car drivers and location scouts and so on to film and edit just a handful of variants let alone thousands.
Maybe there will always be a creamy top layer of increasingly-expensive artisanal handmade advertising but I predict we will end up with a huge sloppy middle ground of generated advertising that is just there to flash bright colours, jingles, movement and brand names into your brain.
I'm not saying it will be good advertising, very much the opposite (not that I think most much non-AI advertising is "good", it's mostly repetitive crap) but I think it will be very cost effective.
Maybe it won't be as effective as "real" ads but it'll be a hell of a lot cheaper and getting 80% of the bang for 5% of the buck means you can do a lot more of it in more channels (or pocket the difference). Every penny you save is a penny you can use to bid for the better slots.
On the graph of spend over the spectrum between that to a genuinely creative live-action advert that is actually memorable for being real (maybe the guy doing the splits between two Volvo™ lorries?) there is a lot of area representing of dross that can be replaced by minimal-input advertotron output. For example 100 million TVs and radios playing in the background while embedding the actual advertising payload of "did anyone say just eat?" into 100 million brains.
Come on, you must have seen a delivery food ad recently. Did the protagonist really have food in their hand or was it AI? What were they wearing? What model was the car in the background? Who cares, that wasn't the purpose of the ad.
Obviously if a creative is bring hired the hiring manager will want to have the best creative they can have for the same money and have the applicants compete with each other for it. But the company board would rather still just not employ that creative in the first place if all they're going to be doing is boilerplate forgettable delivery vehicles for the brand name and you can get 90% of the filler content for that to pop out of your enterprise tier adverts as a service subscription for $50 a month per user.
Which lines up with the rest of what you say that if it's just about hammering the recognition into your grey matter, it's not especially important if the hammer is gold plated.
Businesses presumably spend billions on things like office carpet too and very few of them care exactly what neutral-ish colour it is.
Nearly everyone uses pens daily but almost no one really cares about them or says their company runs using pens. You might grumble when the pens that work keeps in the stationary cupboard are shit, perhaps.
I imagine eventually "AI" services will be commoditised in the same way that pens are now. Loads of functional but faily low-quality stuff, some fairly nice but affordable stuff and some stratospheric gold plated bricks for the military and enthusiasts.
In the middle is a large ecosystem of ink manufacturers, lathe makers, laser engravers, packaging companies and logistics and so on and on that are involved.
The explosive, exponential winner-takes-all scenario where OpenAI and it's investors literally ascend to godhood and the rest of humanity lives forever under their divine bootheels doesn't seem to be the trajectory we're on.
Then again only thinking of fixing things when a customer complains is a way to end up with a leaning tower of hacks which eventually ossify and also the customer (or rather the users, who may not be the customer especially in business software) may be putting up with dozens of niggles and annoyances before they bother to actually report one bug because they can't work around it.
For ads especially no one except career ad-men give much of a shit about the fine details, I think. Most actual humans ignore most ads at a conscious levels and they are perceived on a subconscious level despite "banner-blindness". Website graphics are the same, people dump random stock photos of smiling people or abstract digital image into corporate web pages and read-never literature like flyers and brochures and so on all the time and no one really cares what the image actually are, let alone if the people have 6 fingers or whatever. If Corporate Memphis is good enough visual space-filling nonsense that signals "real company literature" somehow, then AI images presumably are too.
The Londoner is completely out of luck if their seat is a safe seat but not their party.
Not that German politics isn't pretty hosed too.
"Special offers" also really fuck me off. If I want to buy, say, peanut butter today, there's a statistical upcharge because I'm not carefully synchronising my purchase to a secret schedule of when it will cost 20% less. Whenever I see a special offer now, my immediate thought I'd not what it used to be ("wow, that's great"), it is "ugh, these scammers and their constant games".
Charitably, you could call it an elaborate game to make things cheaper for people with less money who can spend the time, privacy and energy (because poor people always have time and energy...?!) to get the best prices. Realistically it seems like mind games to get you used to overpaying most of the time and overbuying some of the time and handing over the data always.
My guess: if the prices reflected the marginal costs of the product inputs, amortised machine wear and ingredient storage and handling and the labour to make, black coffee would be so much cheaper that it would attract too many people away from higher-priced, higher-margin options.
Prices are based on analysis of the effect on demand rather then as a representation of the cost of the item to produce relative to other products.
McDonald's for example has cheap black coffee, because it's an incentive to get you to buy some overpriced food at the same time. Whereas a coffee place is primarily just selling the drinks.
It's missing the Eye of Terror, which should be just below the Crab Nebula.
"All things are poison, and nothing is without poison; the dosage alone makes it so a thing is not a poison."
Internet akchually: a small amount of the heat will be dropped in the wiring that forms the short circuit rather than in the internal resistance. So the panel will be slightly cooler than you'd expect for an object of that colour even in short circuit. They're the same temperature in open circuit, though. When operating normally, they can be quite a lot cooler and in fact you can detect non-functional panels by looking for hot ones with thermal cameras.
And in this context it's much more obvious that it can notionally deliver energy at a peak rate of about 1% capacity per hour. If you said 1MW/360GJ, I don't think that would be nearly as clear.
Same for batteries, which started with car batteries/deep cycle batteries, rather than AA batteries, which usual don't even say, and phones. A battery that provides 1 amp (at 12V, but that's already given in a system) for 50 hours. Makes immediate practical sense, especially when equipment is often labelled in current draw and you can measure amps with an ammeter. 2.16MJ far less so.
Then again, the same also goes for the other storage methods as the spread compresses. Eventually, as always, it all comes down to who can do it on the thinnest shoestring. 1/4 the cost, but the thermodynamic efficiency is 1/3 (direct heat vs batteries + heat pump, say) is still a winner. Finns aren't going to stop needing heat in winter soon, and if you can provide it even a fraction under the cost of battery electricity and a heat pump, you get the customer. And the district heat infrastructure probably already exists.
The problem is that it's not a ecosystem of vast abundance, just occasional abundance. Literally no-one in the world right now is sitting on a constant supply of TWs of excess electrical power and saying "golly gee what are we going to do with all this". Perhaps France got closest in history and their prices still aren't "too cheap to meter".
You can "waste" the power (either by actually "burning" it to heat and dumping it, or just disconnecting the solar panels), but then you'll be short of power later and need to fall back on something expensive or with high externalities. It's also bad in terms of the capex for the solar panels (assuming solar), as you can't use your expensive plant as much as you want. If you can you'd rather use "$10" of energy that you can't sell to store and sell it later, even as heat, at any price than just lose it all.
Even if you massively, massively overbuilt solar and wind so that you were in a "vast abundance" scenario on average, you still have to store some of it for night and/or winter.
In fact even Britain doesn't often use such units - even gas boilers, heat pumps and some AC are specced in kW. You do see BTU (the /hr is often missing) sometimes on AC marketing. My theory is 18000 sounds big and impressive and 5.2kW sounds "meh". We definitely don't talk about "tons", and we buy gas in cubic metres or "units" (which is just kWh)
The coefficient of performance for a heat pump from with cold/hot of 10C/40C is about 10 - this is ballpark where a domestic heating heat pump sits - this is why a heat pump house needs to be well-insulated to work - the heating loop isn't actually hot-hot like with a boiler). From 10C to 500C, it's 1.5. That's 9/0.5 = 18 times less advantage.
At any delta, you can eke out a bit of advantage as COP is always over one, but at some point overheads in the system start to overwhelm your theoretical advantage and you might as well keep it simple and use a cheap and reliable resistor to heat things for dead-cert COP of 1.
You could use a heat pump with a higher COP if you have a really gigantic tank that you only heat to about 40-50C, but obviously the thermal transfer from that is pretty bad and getting it where it needs to go while still being hot enough to be useful is a problem.
If it were pure silica sand, you could presumably get even hotter before anything changes chemically, but at the that point you start having materials issues with metal parts of the system: 500C is about the limit for ordinary steels to lose strength (and many are less than that - heat effects can often start at 300C).
So quite a bang - allegedly this is 200lb, so about the same: https://www.youtube.com/watch?v=ZDgvar7ON54
Insulation isn't such an issue with sand because sand itself is fairly good insulator and obviously doesn't convect. 1m of sand is about the same as 10cm of air. 500C through 1m of sand if roughly 125W/m². Which isn't nothing but it's also 7m from the center to the edge, and the efficiencies only improve the bigger you make the silo.
Presumably they have a double-skin gap and other external insulation too. As the Icelandic hot water pipe systems show, which drop only a few degrees C over hundreds of kilometres of pipe (and thus a gigantic surface area to volume ratio), you can have really quite good insulation if you have space to make it thick.
The hassle of handling hot water is also presumably why they use hot air rather than water as a working fluid for heating the sand in the first place. The worst case if you spring a leak in a heat-transfer tube inside the tank is that a bit of air escapes. Leaking super-heated high-pressure water or steam into the (unpressurised) tank would be a much larger problem, and unloading up to 2000 tonnes of hot, damp, sand to plug it would be operationally very annoying if nothing else.
* Strayed from reality
* Strayed from the document and is freely admixing with other information from its training data without saying so. Done properly, this is a powerful tool for synthesis, and LLMs theoretically are great at it, but done improperly it just muddles things
* Has some kind of bias baked in-ironic mdash-"in summary, this ruling is an example of judicial overreach by activist judges against a tech company which should morally be allowed to do what they want". Not such a problem now, but I think we may see more of this once AI is firmly embedded into every information flow. Currently the AI company game is training people to trust the machine. Once they do, what a resource those people become!
Now, none of those points are unique to LLMs: inaccuracy, misunderstanding, wrong or confused synthesis and especially bias are all common in human journalism. Gell-Mann amnesia and institutional bias and all that.
Perhaps the problem is that I'm not sufficiently mistrustful of the status quo, even though I am already quite suspicious of journalistic analysis. Or maybe it's because AI, though my brain screams "don't trust it, check everything, find the source", remains in the toolbox even when I find problems, whereas for a journalist I'd roll my eyes, call them a hack and leave the website.
Not that it's directly relevant to the immediate utility of AI today, but once AI is everything, or almost everything, then my next worry is what happens when you functionally only have published primary material and AI output to train on. Even without model collapse, what happens when AI journobots inherently don't "pick up the phone", so to speak, to dig up details? For the first year, the media runs almost for free. For the second year, there's no higher level synthesis for the past year to lean on and it all regresses to summarising press releases. Again, there are already many human publications that just repackage PRs, but when that's all there is? This problem isn't limited to journalism, but it's a good example.
China developed and built many UHVDC transmission lines to deal with it.