ChatGPT Stole My Jobs -- Just Like That. And They'll Steal Yours, Too

By Catherine A. Fitzpatrick

So, like everyone, I first fooled around with DALL-E and Midjourney (on the sample version) and didn't do very well myself, and saw that what it really was all about was: a) the prompts -- what flowery and detailed directions you give them and b) your ability to edit it afterwards. Both a) and b) are functions of actual artists; that would not be me. So I tabled it as an issue, although I was more than willing to hear my artist friends complain about it. I didn't especially like the AI art that some of them began doing, which I felt was kind of in the "fractal art" category.

Full of Fakes and Errors

When I first tested ChatGPT, it couldn't generate either my RL or SL biographies because I'm not famous or important enough. When I got it to generate a review of a musician friend, it attributed to her a song she had in fact never written. Where did it get the idea?! It sort of sounded like something she could have written -- and here's where the mobius strip of reflexivity enters in -- now she's going to actually write a song with that title, and even sampled a bit of it at one of her shows. 

I got it to write the history of the Battle of Ilovaisk -- it was credible, accurate, but rather dull and didn't really center the most important point: that Putin offered a "green corridor" for Ukrainians in a kettle but then, of course, was lying from the start, and they were picked off like sitting ducks the minute they lay down their arms and attempted to use the corridor. 

I had it write a review of Second Life -- it was adequate, not going to win any essay prizes.

I tried it again a few times on my own biographies -- finally it ginned them up, likely because there is actually a "mechanical Turk" behind the machine, real people goosing it when it sees Chat stuck. It then rendered a biography of me that simply made up some of the works I had translated, claiming I had translated authors I in fact had never even read much, let alone translated, in one case, and in another, had actually translated, but Chat had no way of knowing that because the information was not public.

My sense of Chatty is that it is scraping not from the wide world of Google, but only from Twitter. Or only from set of stock data bases somebody made up that aren't updated or corrected in real time. Nothing much is ever said about its sources (I have seen articles saying it doesn't access all of Google), but clearly it doesn't have the wide world to scrape that Midjourney and Stable Diffusion have. How long before Google itself puts in the equivalent of robots.txt to stop other companies from poaching on its turf?

Yeah, I get it about the "training" -- which also, to me, gives the lie to any notion of "artificial" intelligence. It's scientists and engineers, using tech themselves to get power over people, then claiming that tech is neutral. It isn't.

So here is how I lost two gigs to the damn thing.

Web Site Copy

The first few gigs with this company involved writing happy copy for the client's web site, and also in one staggering session, writing a humongous marketing document that also had tasks and equipment specs in it. It was in a field with which I was not THAT familiar, so that I had to look up a lot of stuff or ask what it meant. Hours were spent with the client while he tried to explain his ideas. He wasn't a very good writer; he, like a lot of millennials, he never learned to write in school or anywhere else, and mainly texted or chatted on his cell phone.

He had trouble organizing his thoughts and getting them down in a coherent way. He would sometimes be able to get a good outline started, but couldn't seem to work it or it would take forever. So the way he found that what worked best with me was to get me on the phone, talk about the thing, and then I would jot it down in real time on a screen in front of him in Google docs, and then we would both edit it. I would then work it some more on my own, and turn it into him for tweaking. Needless to say, this was time-consuming. At first I charged my regular rates; I ended up not doing this because of the time involved.

I churned out a number of these pieces over some time; some of them brought in more business as clients were very happy; some of them didn't seem to do anything; but since none of them had any negative impact, we kept going.

The client then mentioned to me one day recently that he had heard of ChatGPT, he tried it out with a task he had involving writing descriptions of both a company and various activities -- and was very pleased with the result -- and used it. Oops, there went one of the jobs I would do for him -- just like that.

I cautioned him that the problem with ChatGPT is that it can spout nonsense; it can make up stuff. It puts falsehoods into biographies, reviews, scientific papers as I had witnessed myself or seen others describe. Especially if you had paying clients, they would expect your copy about them to be factually correct and not bring them lawsuits. You had to be really, really careful! Maybe not use it?

How About a Lesser, Low-Paying Job?

Whereupon my client asked me if I now wanted the lesser job -- a much lower-paying job -- of vetting ChatGPT's rendering of a web site. I started looking at it -- and then said "No." It was a painstaking and boring job -- worse than fact-checking a news article others have written, which I have done numerous times, or copy-editing someone's news article, which I have also done endlessly.

That's because you know you are dealing with someone who doesn't care about the facts, feels no sense of liability, and will just say what they need to say. You have to become super vigilant, and also do a lot of your own research, if you knew nothing about the client or subject. It's almost like you'd have to research and write an article or write web site copy yourself, in order to vet ChatGPT's work. No thank you.

I suggested the client could ship this chore to Nigeria -- where he once had had a very competent copy writer for web sites, or Greece, where he has a good marketing guru generating ideas. Argentina, Venezuela, and Vietnam are the countries he sends photo and film editing -- people are desperate for work, and you can find very talented people willing to turn over jobs quickly for less. This is the global economy, and there's nothing you can do about it, really.

What I think may have happened is that no job got shipped. That's because nobody really sits and reads reams of web site copy. Like a lot of things in life, it's there to provide the "look" of something professional; it's filler; it's there to serve SEO in Google searches mainly, and get people to the page to see the offer as such -- everything arrayed around it is really fluff. That's why ChatGPT is perfect for this job. If some error crops up -- eh, you'll correct it.

My client could let it go, and not worry if something fake was in it, because basically, the job of providing fluff was accomplished, with superb SEO to boot.

YouTube Translation

The second gig was translation of a video for a YouTube training session. Translating videos is something that I have made into a certain specialty because I have had to do a lot of it. It can take twice as long to translate a video as it does to watch it, because you often have to go back and replay something. I don't mind the chore.  You often had to make the translation fit as sub-titles on the relevant screen -- something that is always hard to do with Russian, because it is about 75% shorter than the English translation needed to render it. The Russian language has no definite articles and other parts of speech as we know them, and instead are simply attached to a verb or participle or inflection in the voice.

I got into practice doing that shoe-horning of translation copy working with a caption-maker program for TV where we had to take Russian TV and squeeze it into American TV, dropping stuff along the way. Occasionally a grandma in New Jersey might call and complain about it, but it was, as our boss often said when we pressed "send" on the broadcast, it was "good enough for government work" (in this case, city public TV).

But this client's gig, oddly enough, was a task to go from English to Russian, which isn't the norm for me; usually the source language is your non-native language, and the target language is your native language. I can do it adequately, but don't like doing it and it takes longer. The reason for this was that it was a real estate video with the agent providing training to others and describing the meaning of English terms for the real estate business, some of which simply don't have a ready equivalent in Russian. That's because of the years of Soviet state control of land -- although today you find some English terms are now incorporated into the Russian language "as is". So this had the added annoyance of having to decide whether you used the Brighton Beach Angrussky for this purpose, or the actual formal term, i.e. in Russian, "real estate" is literally "non-moveable thing".

Done and Done -- And You're Fired

So I was starting to fuss with this silly thing, when all of a sudden the client came back and said it was finished. He had fed the English into ChatGPT, and it instantly created a very credible Russian text. This text I was willing to take the time to vet at least for a few rounds -- all the terms were correct. I didn't volunteer to check the rest of it because, as Loren Feldman's tag line in his podcast explains, "It doesn't matter."

No one will care. Russian emigres attempting to get real estate licenses need to learn the terms in English, and if they have only a vague idea of the Russian equivalent, well, "good enough."

ChatGPT is just the ticket for the generation that grew up reading "nothing bigger than their hand."

But imagine! I was screwed out of two jobs by ChatGPT, just like that. I didn't even know it could translate. Did you? When did it learn to translate??? It was better than Google Translate -- I checked. 

In this case, the loss was palpable, but not devastating -- this client is the kind of person who, if I am starving, will buy me some groceries or loan me $100 on PayPal, unlike most other cruel employers. 

Why???

This situation came about because children who were never taught to write didn't like writing, and made a thing to do their writing for them. They also were never taught foreign languages, and added that, too.

Why? No one valued it -- or gave up, in the face of the first TV-saturated and then Internet-saturated generation, and here we all are. I can see this not only in general; I watched it up close and personal with my own children, despite my best efforts, and their friends, and I have to say that the difference in ability really has to do with how long a child was able to stay in Catholic school, with its old-fashioned methods, and how good their public school was -- and that decides everything. 

There is absolutely nothing that can be done about this because it's too late; you can't isolate children from the culture they are saturated in, or remove the mobile phones from their hands.

Jesus, Mary, Joseph

Fifty-six years ago, I sat in Sister Sylvester's fifth grade class with my marbleized collegiate notebook, pen in hand, poised for yet another composition lesson.

As this was Catholic school, we always had to start by writing "JMJ" on the top of the page. This stood for "Jesus, Mary, and Joseph". Apparently the Holy Family, versus the Holy Trinity -- preserved for higher functions -- was good enough to get grade-schoolers to focus on the task at hand. They were to provide the inspiration for the job and make sure it got done. Even Catholic schools today have dispensed with this quaint practice, but it never hurts to think of some inspirational or higher power, no? 

Next, came the title of the essay -- this would set the scope and tone and really complete the job for you, if you got it right (which is why editors always change journalists' article titles because they never do the job right). 

So you might put: "The Foundation of Bolivia's Economy: Manila Hemp". Or you might write "Manila Hemp: Powerhouse of the Bolivian Economy" or whatever you managed to devise, as a 10-year-old, with a nun's pointer at the ready to rap your knuckles.

Next, came a big capital Roman number I. This was your main thesis -- this was the "say what you're going to say" out of that magic trio "say what you're going to say, say it, and say what you've said".

So here you might say: "I. Manila Hemp:  40% of Bolivia's Domestic Revenue".

Then there would be A, B, C, where you would explain that Bolivia had the perfect climate conditions for manila hemp and through innovative irrigation practices and ample population for menial labour had caused the crop to flourish (let's say -- I'm making all this up and if it was ever faintly true 50 years ago, today, in fact it's not true, just as a fellow classmate once made up an entire fake essay on Ceylon's main export -- manganese -- and got a "D" on the paper. His outlining technique was excellent, however).

You might describe how the hard-working people of Bolivia adapted to adverse climate conditions (we didn't know about "climate change" in those days) and persisted.

In II, you might describe the actual facts of the crop's component in the economy, citing sources, elaborating, etc. What are the uses of manila hemp? Well, rope-making is one of them. 

In III, you would summarize what you've said and analyze it in some way -- Bolivia was now poised to move into foreign exports; the crop, while not expanding rapidly, appeared to be a stable source of income; migration to big cities by young people might create labour problems. OK, this wouldn't be the scope of Fifth Grade, but you get the idea. And frankly, I believe we took on more in Fifth Grade  back then, than what children tackle today.

I will never forget walking into my son's Fifth Grade public school classroom and seeing a sign that showed all the ways you could spell "dolphin". They were all equal. They were all good. We shouldn't judge any of them. Children generated them out of their little brains; they got "A" for effort. With that brand confusion around the word with an "ph" -- in the absence of any phonics teaching whatsoever -- I wonder if the kids ever did learn to spell dolphin. Probably not.

Like diagraming sentences -- another Catholic school forte still taught today -- these lost arts weren't really valued or needed, so they atrophied.

Google Translate Didn't Take My Jobs - Why ChatGPT?

Google Translate did not put me out of a job as such, because I translate lots of different kinds of things -- books, catalogues, articles, memoirs, historic letters, legal documents, and generally I'm always going to be better -- and clients and publishers and editors know that.

But at one time, bread-and-butter translation work would involve simply writing letters for people or translating the letters they received back, in business and in personal matters like bride-buying abroad (after doing a few of those gigs, I quit because they were so horrible and exploitative -- often by both parties). Today, that sort of thing is accomplished more or less with Google Translate or Yandex (which is a lot better). I myself will pop a text in Google I need to write, translate it roughly, then take it to Yandex to type in what I know to be accurate. I will never help Google because I don't believe in doing unpaid work for tech behemoths.

A most despicable practice of Google is to go to a place like Kyrgyzstan, and recruit a lot of eager young people looking for high-paying tech jobs who have learned English, and get them to help fix up Google translation of their language into English for no pay. The tech world often treats translation as a subbotnik and won't accept that it should be paid like their programming is paid.

But ChatGPT instantly took me out of a job by having a better interface; by being willing to take longer texts in its box; by having different AI functions at work, I guess. I don't know the mechanics, but clearly it's a genius. Sure, we should see how it does with Tolstoy and Dostoyevsky, but few people are going to pay you to retranslate them; the bulk of the jobs today are in news, technical, and legal translation. I get the feeling ChatGPT isn't so much fetching and carrying as "translate" literally means, but is manifesting the inherent English in a Russian text -- if that makes sense. Different.

Goat Herds

I have a vivid image before me of Elon Musk's demo of a robot recently, that was put to work on a Tesla assembly line. It lumbered out to the shop floor and went to work picking parts out of crates, then setting them up and assembling them on a conveyor belt.

But nearby was a man with a big hook like a goat herd. The robot would sometimes wander out of its designated path or get stuck or even start to fall over -- it's top heavy. So the robot herd would prod it back into place with the big stick.

And that's the kind of job that will be available to us with ChatGPT, which is yet another kind of robot, a robot in the hands of humans who want to take power over other humans and cloak it with the terms "fun" and "science" and "enterprise". You'll get to poke and prod the machines, being careful not to get stepped on yourself.

One of the jobs I saw go by on Upwork (which is an atrocious, low-paid slave labour place) or some other job ads service involved literally going over AI-translated texts and fixing them up. And not even with the full, coherent text in front of you; you would work with a huge stream of CAPCHA like nonsense texts.

Foxes in the Hen House

There is a tsunami of articles generated by tech journos who are totally craven to the industries -- they amount to payola. They assure you that ChatGPT (probably) isn't going to take your job. That's not how to cover this subject. It's like having foxes assess how secure the hen house is, and how well the hens are going to do in their presence.

Rather, send out queries about people who already lost their jobs, full-time or contract gigs. Fact-check the self-reporting with their employers. Ask people how they used GPT and see if it took away a job. Of course it did, and there will be more.

Right now the free version is over capacity and cannot be used. So I tried one of its competitors, Character.ai which has a persona to ask your questions from. The results are terrible and don't reflect the nature of the historical characters at all.

I can't see any mitigating factors here. That the free ChatGPT is "down" and can't be used is meaningless, as people will pay for it. That these Characters are silly and stupid is meaningless as people might even pay for that entertainment, too, and it will get better. 

Meanwhile, I assure you I will go on writing my blog posts myself, without robots, and you will go on not reading them.