“Why does that statue of Abraham Lincoln have 17 fingers?”
I’m the web developer for a company that sells used industrial machinery. I work closely with the guy who takes care of all the marketing. He’s always used a mix of sources for marketing copy - sometimes he writes it, or has the salespeople write it, or he has one or another marketing company write it.
He’s super in to ChatGPT now. Not gonna lie, he has been using it for everything lately. Sometimes with good results, sometimes with not-so-good results, sometimes with no results at all. Feels like he’s using it as a replacement for Google almost.
Yesterday he sent me some copy for a new advertising page and I could tell it was from ChatGPT. Because it had good grammar and was wordy. But still it didn’t feel like it had any DEPTH. Granted, a lot of marketing copy seems shallow and buzz-word-y. But I think text from a real human marketer would have been more pleasant to read.
It also was just a wall of text. I had to go through it and pull out some phrases to use on the rest of the page, and a phrase to use as a title, because ChatGPT didn’t provide any of that. And, I still don’t know if the text was very well “keyword optimized.” When we get assets from the actual marketing companies they come with stats about the keywords they used and why they used them. This text just kind of had the same words and phrases over and over again.
IMHO ChatGPT is not ready to roll on its own right now, but I think people will be using it. I don’t know if this will make its output better or worse. Like how videos from your phone are actually very crisp and HD, but as a video goes viral and gets shared over and over again it gets more and more compressed and ugly as it goes through more sources. I don’t know if ChatGPT putting out crap content that people use will then get sourced into ChatGPT to create even crappier content that people use ad nauseum.
It will still take manpower to initiate the content, check the content and then put it in to production. But knowing what we do about how capitalism works, people are willing to put up with crap if it’s cheaper crap.
I say there will still be room for marketing professionals, you just need to position yourself at the top of the heap so you can provide the services that ChatGPT can’t provide, or that give ChatGPT usefulness.
If human-generated texts (and images) are gradually replaced, in many realms, by AI-generated ones…
…but AI draws from human creations to cobble together its output…
…won’t AI eventually start running out of new human-generated input from which to draw?
At some point won’t it start drawing from (recycling) earlier AI-cobbled stuff?
So, creative human input (and therefore AI output) will be frozen in time, to the early 2020s. A little like how, on distant stars, life forms might get to enjoy “I Love Lucy” and “MASH,” but not broadcasts from before, nor from afterward — just that window of a few decades of human history.
The current models are fed on everything from cave paintings to recent contemporary works so there’s a lot of stuff for it to draw from. At this point it’s more about creating models that can turn that data into the images you request. As long as I can keep coming up with new ideas, it should be able to keep crafting unique works with the correct prompting: “I want a painting of a vase of flowers carried by a camel sort of in the style of this artist but also in the style of this completely different artist but with more of a pastelpunk vibe and looser strokes but also it should be in a medium like spray paint applied to wet cardboard”.
Also, I don’t foresee a time when humans stop creating so it should always have some source of new images.
People would rather have Wikipedia for free and accept it is full of crap than pay money for Encyclopaedia Britannica which is supposed to be fact-checked.
So when the AI marketer lies to the marketing project buyer with the intent of selling more marketing projects that’s bad. But when the marketing materials the AI creates lie to the end-customers about the end-product with the intent of selling more end-product that’s good.
Not calling you out; you just brought into clear focus the essential conundra of all middlemen and of all marketing. Namely that lies are good until the person being lied to both notices and objects.
Nothing new about that. 30 years ago we were selling a design automation tool and we had a real-life carbon-based person write the press release. ChatGPT could have done much better. “Product works by algorithms.”
A lot of marketing writing is basically mindless.
BTW a few years ago I heard of software that basically wrote descriptions of local minor league baseball games based on box scores. It was all so regular that a human seldom needed to be involved.
The real danger of “artificial intelligence” (even pretty limited ‘generative’ AI) is that the convenience will encourage people to pick the cheap and easy option rather than to do the work to develop their own skills, knowledge, and ‘voice’. Nothing that ChatGPT or Bing any of these other engines—and they are just machine synthesis engines, not actual autodidactic intellects—is in any way original, although lacking the constraints that most people have in terms of their cognitive constructs it often appears superficially novel and more importantly can be produced in a fraction of the time that a writer can write an essay or an artist can produce an image. Since much of what is produced as commercial art is derivative anyway, it looks and feels equivalent (at least, to the unstudied eye) to human-produced content but it really just serves to erode interest in developing the skills and knowledge base to do this work in a manner analogous to how the electronic calculator and then computer algebra systems have caused a decline in skill of basic arithmetic and complex mathematical manipulation by all except an obsessive minority.
When we do actually produce a genuine artificial general intelligence (AGI) that is capable of original work, the output will probably not even make sense to us except in a very trivial manner of direct application. That is to say, if an AGI developed into the equivalent of Federico Fellini or Martin Scorsese, it would produce the analogue of a ‘film’ that wouldn’t even make sense to human viewers, and of course we will not have any insight into how it functions ‘under the hood’ even at a granular level. Fortunately, despite all of the hype, I don’t think current ‘chatbots’ are anywhere close to this kind of breakthrough, and are actually probably a dead end toward genuine AGI. Which is not to say that they won’t have a dramatic impact upon intellectual workers, and that people currently producing art and other mind labor content aren’t going to have to learn to work with and utilize these systems because they are being adopted writ large as fast as people can figure out an application and appropriate training set, regardless of whether this is really a good idea or that they are as reliable as advertised.
It seems like a lot of people aren’t aware of it but there is a vast amount of online content that is automagically generated and has been for about the past decade. Given a few basic prompts and a selection of images, you can put up an automatically generated YouTube video on any topic that is…well, annoying and repetitive, but at least as good or better than most of the content uploaded by actual human contributors. The only thing special about ChatGPT and other chatbots is their near real time interactivity, which is possible through the enormous amount of computing power that can draw from a vast array of ‘trained’ data and the connections made between different elements via the complex neural network that is weighted through extensive training. This is emergent and in a sense represents a breakthrough of sorts, but really one of complexity versus actual capability. Whether you want to argue that a chatbot ‘understands’ what is asked of it is something of a semantic argument, but they are now sophisticated enough to respond to a pretty general prompt with a product that approaches being useful. And as you note, marketing blurbs and the like contain very little semantic information, anyway; they’re basically a way of packing some small amount of useful data into an aesthetically pleasing package of words, images, and music depending on the format; it is basically the perfect application for a generative AI.
Stranger
There are prompts you can give it to make it look more human-like in the output (and defeat all those is-it-real-or-is-it-AI tools out there) and you can certainly ask it to trim down the copy, change the tone, etc. There’s a bit of an art to engineering a prompt or set of prompts to get it looking more natural. I do agree that it lacks significant depth or bursts of inspiration.
(This is a prompt that works for making output a bit more human:)
As an experiment, I asked ChatGPT to write 10 email subject lines about a non-work topic. They were totally bland business speak. Then I wrote, “Make them funny,” and the results were bonkers. Some made no sense, but others were Onion-caliber.
Yeah, you have to do some human curating and prodding to lead it along to the destination you’re hoping for, if you want it to be more than fairly generic prose. Humor prompts I’ve found to be pretty hit-or-miss, but I usually ask for 5 or 10 variations, and there’s usually one or two that are pretty good and actually make me laugh. One of my favorites that I’ve mentioned before was looking for an ominous one-liner to end a statement about AI and the coming singularity, and it mostly gave me bland stuff, sans one: “let’s hope the machines are more forgiving than the humans they replace,” which I thought was gold.
Wiki is not full of crap. If you start looking around the internet, you’ll find that it is surprisingly good, a lot of times better than Britannica, and a lot of times worse. To just dismiss it like that just means that you’re not up to date on the issue.
Wikipedia is acknowledged as the best source of information online for knee arthroscopes, for example. Its cancer information is as accurate and in-depth as a database maintained by experts. Its nephrology information is comprehensive and fairly reliable. Its drug information is accurate and comprehensive, even when compared to textbooks. Its political coverage is accurate. It’s a highly complete and accurate resource on musculoskeletal anatomy.
Maybe the biggest surprise of the internet to me is that false bubbles generally are corrected on Wikipedia. Even if you pick a controversial topic like climate change or vaccination, the Wikipedia page typically reflects mainstream scientific consensus.
I once asked ChatGTP for 25 captions for New Yorker comics (I was really hoping it would write “Christ, what an asshole” 25 times.). Almost all of them were bland or bad. One was really, really good.
“The perils of online dating in the city.”
“Just another day in the life of a New Yorker.”
“The struggle of finding a decent apartment in the city.”
“When you thought you were having a bad day, but then you see someone wearing a cat on their head.”
“The never-ending battle of avoiding tourists in Times Square.”
“The joys of commuting on the subway during rush hour.”
“When you accidentally make eye contact with a stranger on the street.”
“The never-ending search for good coffee in the city.”
“When you finally score a seat on the subway, but it’s next to a smelly person.”
“The endless search for a decent slice of pizza in the city.”
“The joys of being a New Yorker - always on the go and never stopping.”
“The battle of trying to hail a cab in the city.”
“The never-ending search for a good bagel in the city.”
“When you thought you had it all figured out, but then the city throws you a curveball.”
“The daily hustle and bustle of city life.”
“When you thought you were alone in the city, but then you see a rat.”
“The never-ending quest for a decent cup of tea in the city.”
“When you finally find a parking spot in the city, but it’s in a no parking zone.”
“The joys of trying to navigate through the crowds in the city.”
“When you finally score a table at your favorite restaurant, but then the service is terrible.”
“The never-ending search for a good bagel and lox in the city.”
“When you thought you had a good idea, but then the city proves you wrong.”
“The joys of trying to get a good photo in the city without any tourists photobombing.”
“When you thought you were having a good day, but then you step in a puddle of water on the sidewalk.”
“The never-ending search for the perfect cup of coffee in the city.”
I assume you think there was a good one in that list of tiresome banalities? Or were you going to post the good one in a different post after we all go nuts trying to find the nonexistent need in the small haystack?
Sounds like a list of episode titles of a Friends/Seinfeld-style 90s sitcom.
I wish I was.
I suspect that a lot of companies will end up trying AI for this sort of thing long before it’s ready, and will get cheap crap. It’s like the stuff you see in “articles” online that are obviously written by AI programs. They suck, but they’re
- Cheap, and
- Good enough to serve the purpose of SEO, which doesn’t require the article be good.
Eventually, it’ll be apparent the work of AIs isn’t good enough for a lot of things. Manufacturing went through a similar process when cheap stuff from China became easy to get; it seemed to make sense to buy everything from there. A LOT of those things are being bought in North America again because dealing the with the low quality often just isn’t worth it. China is good for some things to be sure, but not others.
Sounds like a better success rate than actual New Yorker comics.
Um, I thought it would be obvious?
“When you thought you were having a bad day, but then you see someone wearing a cat on their head.”