AI is upending the freelance world | Hacker News Reader

AI is upending the freelance world

forbes.com

145 pointsgeox160 comments

This is kind of baffling to me. I'm a huge fan of the usefulness of ChatGPT, and I've found ways to use it to improve my productivity in a bunch of areas, but if you've "never hired a writer better than ChatGPT", I can only assume you've been hiring really shitty writers. Not necessarily that ChatGPT output is bad, but a lot of the copy has a similar, generic, "college essay" feel to it.

I saw this post on LinkedIn [1] that I think really nails it:

For anyone worried about how generative AI will make creative jobs obsolete, I asked Chat GPT to generate headlines advertising hot cross buns. Here's what it came up with:

"Get Your Hot Cross Buns Here - Fresh and Delicious!" "Easter Treats Just Got Better with Our Hot Cross Buns" "Warm Up Your Morning with Our Hot Cross Buns" "Limited Time Offer: Hot Cross Buns Now Available!" "Our Hot Cross Buns Will Leave You Wanting More" "Celebrate Easter with Our Traditional Hot Cross Buns" "Delight Your Taste Buds with Our Soft and Spicy Hot Cross Buns" "Bring the Aroma of Freshly Baked Hot Cross Buns into Your Home" "Satisfy Your Cravings with Our Irresistible Hot Cross Buns" "Hop into Easter with Our Scrumptious Hot Cross Buns"

Meanwhile a human copywriter just created this real headline for a press ad for Tesco hot cross buns:

"Go Out With All Buns Glazing."

If you want originality, ask a human.

1. https://www.linkedin.com/posts/realtommorton_for-anyone-worr...

Eh, that's a specific type of wordplay and not so clever. Just add it to your prompt.

> a similar, generic, "college essay" feel to it

I worked for a large periodical publishing corp throughout my 20s. This + quick turnaround is exactly what the publishing staff (ad sales) fought for, because it was safe. Only the edit staff (writers, etc.) wanted anything else, and those who really wanted it left.

Sure ad sales people want that, and if you let them, they will tank the publication, and then ad sales will be 0. That's the reason you have the edit staff...

In spirit I agree, but I think you picked a bad example. With just version 3.5 I got these. They aren't cherry-picked, I just massaged the responses with a little back-and-forth (funnier and punnier) Here are the first 5.

Bun-believable! These Hot Buns Will Have You Saying 'Oh Doughn't Mind If I Do'" "Hot Buns: The Yeast You Can Do for Your Taste Buds" "Bake Your Day with These Hot Buns: The Roll-ly Grail of Breakfast Treats" "Raisin the Bar: How to Make Hot Cross Buns that are Bun-derfully Delicious" "Don't Be Alarmed, These Hot Buns are Oven-tually Going to Be Your Favorites"

A compromise is that your prompt has to be creative.. so a human is still in the loop for now..

On the plus side, it probably delivers on time.

I think you missed the point. They quoted an advert involving a pun because it was creative, you are just showing results because it includes puns as a response. Most of those are used or subpar.

Yeah, it's very possible I missed the point. I think though that most publications will settle for something that's good enough. They don't need (or can't afford) Steinbeck headlines, so they'll go the easiest and cheapest route. I didn't personally find the original headline any better than the generated ones, but maybe I don't have as refined a palate as you do. Thanks for the feedback.

These are all very simple low hanging fruit puns. None of these come close to the human example.

That said, I'm not an idiot - AI will get there and probably soon.

unless we encounter a devil's staircase.

I asked ChatGPT-4 to do this, but I added "in a creative, original, and witty way". This is what ChatGPT gave me:

1. "Heavenly Buns on the Rise: Our Hot Cross Buns are Simply Divine!"

2. "Bunbelievable! Our Hot Cross Buns Will Make You Hoppy this Easter!"

3. "Hallelu-bun-jah! Feast Your Eyes (and Mouth) on These Scrumptious Hot Cross Buns!"

4. "Cross Your Heart and Hope to Buy: Irresistible Hot Cross Buns Just in Time for Easter!"

5. "Hot Cross Fun: Get Ready to Bun-dle Up with Our Tasty Easter Treats!"

6. "No Bunny Compares: Hop into Easter with Our Delectable Hot Cross Buns!"

7. "Get Egg-cited! Our Hot Cross Buns are Simply Bun-derful!"

8. "Cross Your Buns and Dot Your I's: The Ultimate Easter Treat has Arrived!"

9. "Bun Voyage: Embark on a Flavor Adventure with Our Heavenly Hot Cross Buns!"

10. "Holy Buns-a-roo! It's a Hot Cross Bunanza You Won't Want to Miss!"

I thought they were pretty good. And honestly better than what I'd come up with. But also this isn't something I'm good at, or even good at judging. So take my opinion for what its worth given my lack of experience here, but I thought it wasn't too bad.

UPDATE:

I decided to ask ChatGPT to do it again, but this time to remove references to Easter and instead make it NBA Playoff themed. This is what it gave me:

1. "Bun and Dones: Hot Cross Buns that Make Your Taste Buds Slam Dunk!"

2. "From Downtown! Our Hot Cross Buns are a Three-Pointer in Deliciousness!"

3. "Courtside Cravings: Get Hooked on Our Sky-High Hot Cross Buns!"

4. "Pick and Roll Your Way to Flavor Town with Our Dribble-icious Hot Cross Buns!"

5. "Give Your Appetite a Full Court Press with Our Swish-worthy Hot Cross Buns!"

6. "Treat Yo' Self to a Layup of Yum with Our Un-bun-lievably Tasty Hot Cross Buns!"

7. "Bunbelievable Rebounds: Our Hot Cross Buns Will Keep You Coming Back for More!"

8. "A Slam Bun-kin' Good Time: Taste the Victory with Our Hot Cross Buns!"

9. "Step Up Your Snack Game with our Ballin' Hot Cross Buns!"

10. "Hot Cross Buns: Bringing the Heat to Your NBA Playoff Watch Party!"

I liked that I could tweak it toward a direction and get instant feedback.

Heh, in the near future openAI devs will wonder why the training set has a wad of bun puns stuck in the middle.

Yup, I'd say that is on-par with the "human headline". Maybe ask to shorten it, and you might get an even better result.

Yeah too long. All of them.

These are 100% better than what I'd come up with.

I describe it's writing style as "8th grade honors student". It's all very correct and perfectly structured but facile and frequently reductive and with obvious errors you'd expect from someone who has only read about the world in books.

It's the writing style that is the most congenial to the base average human reader. To a reasonably smart person it gives that artificial vibe but to the network TV watching masses it's perfect writing.

Depending on the need that may be perfectly fine. I wouldn't expect to be published in the New Yorker any time soon, but it can write press releases, marketing copy, technical documentation and probably a load of niches (real estate listings?) which are predominantly written by non-writers right now.

You can add a New Yorker writers name or ask it to use the New Yorker style guide in your prompt

So slightly more sophisticated than the average American?

Significantly more sophisticated than average anywhere. People are not good at writing. 8th grade is extremely generous

It's naïve to think LLM's won't learn to make trite wordplay based on common idioms in the near future.

Just tried "Can you generate some options for headlines advertising hot cross buns utilizing wordplay, with a cheeky attitude?"

    "Bunbelievable! Get Ready for a Crumbelievable Hot Cross Experience!"
    "Bun in a Million: Your Taste Buds Won't Be Cross with These Hot Cross Buns!"
    "Hopping Mad for Hot Cross Buns: They're the Toast of the Town!"
    "Cross Your Heart, Hope to Savor: Hot Cross Buns Worth Sinning For!"
    "Buns & Roses: It's a Rockin' Hot Cross Bun Party You Can't Resist!"
    "The Great Bun-der: Hot Cross Buns with an Irresistible Twist!"
    "Hot Cross Fun: Unleash Your Inner Bunthusiast Today!"
    "Bun Appetit! Tantalize Your Tastebuds with Our Heavenly Hot Cross Buns!"
    "Too Hot to Handle: These Hot Cross Buns Are a Batch Made in Heaven!"
    "Sweet Cross-sations: Get Ready to Indulge in the Ultimate Hot Cross Buns!"

And more fun

"Now try that again adding a hint of dissatisfaction with the commercialization of Easter while acknowledging that this is itself is a form of that commercialization."

    "Crossed Wires: Enjoy Our Hot Cross Buns While We Roll Our Eyes at Easter's Over-Commercialization!"
    "Eggs-cruciating Irony: Hot Cross Buns So Good, They'll Make You Forget the Easter Bunny Cash Grab!"
    "Bun in the System: We Can't Fight the Hype, but Our Hot Cross Buns Are Worth It!"
    "Commerciali-sin: Indulge in Our Hot Cross Buns and Forgive Us Our Easter Retail Trespasses!"
    "Capitalist Crossroads: Relish Our Hot Cross Buns as We (Reluctantly) Join the Easter Bandwagon!"
    "Hoppin' Hypocrisy: Grab a Taste of Our Hot Cross Buns as We Grumble About Easter's Materialism!"
    "Bitter Buns: Savor Our Hot Cross Delights While Lamenting the Overhyped Easter Frenzy!"
    "Eggs-traordinary Irony: Bite into Our Hot Cross Buns and Pardon Our Easter Commercial Participation!"
    "Rebel with a Bun: Enjoy Our Hot Cross Buns and Share Our Disdain for Easter's Selling Out!"
    "Crossed Purposes: Hop on the Easter Cash Train with Our Hot Cross Buns, and We'll Share a Collective Eye Roll!"

It gives you what you ask for, not what you want.

None of these are anywhere near any good as the human delivered one, for so many many reasons. It's a mediocre list even given what you asked for.

"This talking dog is a complete dumbass. It has no idea how to make a good risotto, and the C++ code it wrote has security holes you could drive a truck through. I don't get the hype."

Unless I've missed a change to the topic, the question isn't how unbelievably remarkable the ChatDog is, but if it's good enough to hire.

From my experience, in software, I'd say it's like a junior fresh from university — it certainly has its place, but humans can still regularly best it.

I'm expecting both that it will be used appropriately to boost productivity and lower cost, and inappropriate by people who don't have enough domain knowledge (or wisdom) to check the results before using them: https://www.google.com/search?q=%22as+a+large+language+model...

It'd be a missed opportunity if someone with a talking dog tried to hype it as a replacement for C++ programmers instead of an entertaining sideshow - no-one wants a talking dog that writes terrible C++ code to be writing their C++ code.

It'd be as silly as cirque du soleil advertising that their amazing trapeze artists will be replacing accountants any day now.

These other guys ignoring how awesome a talking dog would be, fools!

Yes, that is exactly what I'm saying, because the context for your analogy statement is an entire website of my peers (although almost none with as much experience as me) saying that the talking dog can make a perfect risotto, that it's C++ code will replace all programmers, and we'd also better watch out for the global domination of the talking dog's future puppies.

I applaud your creative way of second-level prompt engineering!

It used to be the best way to get Linux support was to join a Linux chat room and loudly proclaim Linux can't do something - ten amazing people would jump up to prove you wrong.

Turns out this is also how to use ChatGPT. :-)

This is the way

Agree! See: Cunningham's Law states "the best way to get the right answer on the internet is not to ask a question; it's to post the wrong answer."

The vast majority of "creative" work is just mediocre, or more specifically, passable.

By majority I don't mean > 50%. I mean > 99%.

"Go Out With All Buns Glazing" represents < 1% writing amoung all the writing-related jobs. And even that is not entirely impossible with ChatGPT today.

Also, employers dont want quality. They want quantity.

Exactly. And speed.

"Getting there firstest with the mostest".

They want quality but cheap, we'll be fun to watch how they complain to an AI if the results don't satisfy.

Most likely they would just run the prompt again with some tweaks. You could ask "generate me 50 slogans for xyz" all day long. Do that with a human and they'd get fed up quickly.

As do freelancers. this has been the understanding for as long as freelancing has existed, that 90% of your mediocre pay-the-bills projects just get you by until you land the 10% of work that is interesting, rewarding, and utilizes your talents the fullest.

AI is going to destroy that bread and butter (har har) lifeline that freelancers sustain themselves on while they search for opportunities which really leverage their human creative qualities in a way that AI can't mimic. But now they can't either, because they can't afford rent.

This is likely untrue. I think tone is something ChatGPT excels at. The issue is more likely with your prompting. Sure, with standard prompting you'd get the "college essay" feel.

However, I do think long form writing is something ChatGPT isn't really able to do.

The way ChatGPT writes drives me nuts in some specific ways. So I was messing around with it and asked it to specifically eliminate certain phrases it commonly uses, remove superfluous words, and modify the sentence structures in some specific ways. In the end it was surprisingly easy and fast to get it to produce writing at a significantly higher than 8th grade level.

Scroll through the emails you haven't bothered unsubscribing from yet

They all sound like that. It's bland uninspired copy - probably with some emojis

Side note that I hope someone at OpenAI is scratching their head over the sudden influx of hot cross bun related queries

> I've found ways to use it to improve my productivity in a bunch of areas

like what?

finding answers quickly, writing bash scripts, exploring ideas & planning, writing app boilerplate

> I can only assume you've been hiring really shitty writers

Which is what 99% of companies hire because they don’t have budget or think they shouldn’t spend more? Talented writers, designers, programmers, lawyers etc won’t be gone soon, but the 99% of crap out there (most of fiverr, most of upwork etc) will be done soon. And while that seems a good thing in some ways, it is a massive economical and societal issue.

The output of chatgpt is ‘good enough’ for most things and for the price (near $0) of it, you can do many iterations yourself without paying a writer.

Well they mention in the article she was paying an average of 22$/hr which is definitely not what you pay for a good writer.

Can you please share from your experience what a "good writer" costs per hour?

Here in Europe easily up to 60€ per hour or based on assignment. Especially for commercial or corporate work. Way more if the writer has an industry-recognized byline.

Entire product ranges are sold from China with terrible documentation, but still sell. Somehow the people who write those get paid, so I assume there’s a massive range of professional skill levels.

I hope Chinese manual writers are currently using ChatGPT for new products. Surely this problem stops now. If not, filling that gap is an opportunity for someone.

> Not necessarily that ChatGPT output is bad, but a lot of the copy has a similar, generic, "college essay" feel to it.

So it's just like the output of these traditional media articles.

So much online "content" is just product sales or thinly veiled SEO fodder. If I were a "content writer" and I wasn't servicing high end clients, I would be getting worried right now.

The bulk of online content creation is for SEO and content marketing. This doesn't require creativity.

It's not going to be a disaster if this content disappears from our culture altogether. Sad for writers though, but every single freelance writer I know does this just to pay the bills and dreams of writing something actually creative.

This sounds like the Kenyan college essay industry conversation the other day. I wonder if the same arguments will apply ("this is good for them because they can do more creative things instead or start businesses to help the local economy", etc)

Caroline O'Donoghue has written about exactly this phenomenom. As a creative writer, you only ever see your "competition's" top work. All the crap that you are forced to do to pay the bills never gets any attention.

Her blog is here: https://sentimentalgarbage.substack.com/p/the-pied-piper-is-...

> "never hired a writer better than ChatGPT"

Indeed. I stopped reading after this inane quote.

This is how the future of communication will look:

    1. Sender writes a bunch of bullet points and asks ChatGPT to convert it to coherent long form text.

    2. Receiver takes the long form text and asks ChatGPT to summarize it into bullet points.

You might have meant this in jest but considering some of our human to human communication is already like that, I think this is 100% prophetic.

This human copywriter pun is extremely bad. I don't think any respectable agency would run with it, simply because it has highly inappropriate sexual connotation that could potentially be offensive to many customers.

Ad like this could easily get someone in trouble when used in wrong context (neurodivergent people especially have a problem with this).

Sadly, in general people do not appreciate good writing. Translators, for instance, have little status, and do not get paid very well unless they deal with legal ol live interpreting.

Can we name the translators of some books on our shelves? Personally, no. I only remember the three body problem translator is the brother if the author, IIRC.

I also gave it a shot. I'm Dutch and never heard of hot buns before, so I first tried to generate the prompt:

* This is a slogan for hot buns: "Go Out With All Buns Glazing." Can you explain why this is a good slogan? "Go Out With All Buns Glazing" is a good slogan for hot buns for several reasons:

* Can you list 10 characteristics of hot buns?

* Can you list 10 characteristics of Americans?

* As a hot bun company CEO can you write a briefing for a marketing firm to come up with 10 new ideas for a slogan. The slogan should contain at least 1 pun. The briefing should be 1000 words long. Please use this information in the briefing . A good slogan has these characteristics: [copy paste of previous answers]

And then in a new session:

As the creative director of a marketing firm, you get this briefing. Please generate 10 slogans. ```As the CEO of [Hot Bun Company], I am excited to collaborate with your marketing firm to develop a fresh, creative, and engaging slogan for our brand. Our hot buns have a reputation for being freshly baked, warm, and delicious, and we want our new slogan to effectively convey the unique appeal of our product. In this brief, I will outline the essential elements we would like to see incorporated into our new slogan and provide some background on the characteristics of our hot buns, as well as some general observations about common cultural characteristics often associated with Americans. Our goal is to create a slogan that resonates with our target audience and sets our brand apart.

To create an effective slogan, we would like to emphasize the following characteristics:

Play on words: Our new slogan should contain at least one pun, making it clever, memorable, and easily relatable to our product. Evokes imagery: We want the slogan to bring to mind the image of warm, freshly baked buns, enticing potential customers to try our product. Energetic tone: The slogan should have an energetic tone that appeals to consumers and makes our brand more memorable. Positive connotations: Incorporate elements that suggest our hot buns are made with great care and enthusiasm, encouraging customers to try them. Short and catchy: Finally, the slogan should be short, memorable, and easy to understand. When creating the new slogan, please consider the following characteristics of our hot buns:

Freshly baked Soft and fluffy Warm Glazed or topped Aromatic Versatile Golden brown Moist Comfort food Shareable Additionally, we want to ensure that our slogan resonates with our target audience – Americans. As such, please keep the following general cultural characteristics in mind during the slogan creation process:

Individualism Patriotism Optimism Consumerism Diversity Informality Work ethic Philanthropy Entrepreneurship Mobility By incorporating these elements, we believe that we can create a slogan that not only encapsulates the essence of our hot bun brand but also resonates with our target audience. Our objective is to receive 10 new slogan ideas that embody these characteristics while integrating at least one pun.

We trust your marketing expertise and creativity to develop a slogan that captures the attention of consumers and highlights the unique appeal of our hot buns. We look forward to reviewing your ideas and working closely with you to finalize a new slogan that will become synonymous with our brand.```

And the results:

"Bunbelievable Freshness: Taste the Warmth!"

"Hot Buns: America's Baked Bliss"

"Rolling in Flavor: Savor the Golden Goodness"

"A Bun in Every Oven: Fresh, Warm, United"

"Rise to the Occasion: Hot Buns, Hearty Smiles"

"Bake My Day: Irresistibly Soft and Shareable"

"Bun Voyage: Embark on a Warm, Fluffy Adventure"

"Bun Appetit: Taste the All-American Aroma"

"Happiness is a Warm Bun: Freshly Baked and Ready to Roll"

"The United Buns of America: Let Your Taste Buds Soar"

Not award winning, but some are pretty good.

"Bunbelievable", "Bake my day", "Bun appetit", "United Buns of America", "Bun Voyage", ""Rolling in Flavor" are all blatantly plagiarized, as a quick Google search reveals. I stopped there.

"A Bun in Every Oven" is rather risky as a slogan ...

In other words, zero creativity and not much beyond a Google search if one is willing to plagiarize.

In fairness, "all buns glazing" is also blatantly plagiarized by the same measure.

An OpenAI plugin should help with that.

But yeah, like another commenter mentioned, most bun-puns are already created..

You need to give it more context. If you don’t give any style guidance then of course you get the default chatgpt style

This thread makes it sound like people are overestimating how good the content is supposed to be. Most clients want whats in their head, in writing, and they do not fully comprehend or care when its written badly (they aren't writers), it should convey the idea that they already have, to their potential audience. I have a client for which I develop software, who now writes 3-4 articles per day with ChatGPT, he no longer works with writers who take at least a day or two to put out one article at a time, and often need revision. And it is paying off looking at ranking and analytics of his product.

Yes, the content is not better than a human writer, but what my client is seeing:

1. Much faster turnaround.

2. He gets exactly what he wants and can revise himself.

3. Better results in his ranking due to the content.

4. Some sort of "Engine" which understand exactly what my client is talking about without getting confused.

Isn't that a "better writer" than a human in this case? In his mind, if he goes back to a human writer, he loses a very important and measurable edge he has gained over the past 4-5 months.

> Now, freelancers who are less experienced and don’t offer specialized skills stand to lose their gigs, according to five clients Forbes interviewed

And even then, I wonder if general writing quality improves, so that clients can't prompt-engineer good enough on their own and need to (you guessed it) hire freelancers to use GPT for them.

You can GPT's "voice" by explicitly prompting it, e.g. to be more concise and use less big words, which makes it nearly impossible to distinguish from a human. And GPT-4 genuinely writes very well, well enough so that (especially after changing the voice) you don't need to hire a writer. Mind you, writing is a task people were getting payed very cheap to do, and I still don't think GPT-4 can write a best-selling novel or blog post all on its own (it can come up with creative ideas, but IMO still not as creative as a human; time will tell).

Give those metrics a year, and we’ll see how it turns out. This reasoning completely disregards the long term vision for a clear quick gain. We’re in a reality distortion field where, for some reason, everyone thinks that the circus trick performed by GPTs is an absolute, it’s here to last forever and it will just get better. I can’t think of any other technology that has fooled business people so quickly into adopting it without ever worrying about the alpha nature of the tech itself. Feels like mass hypnosis

> We’re in a reality distortion field where, for some reason, everyone thinks that the circus trick performed by GPTs is an absolute, it’s here to last forever and it will just get better

Why would it not last forever and improve? Unless OpenAI is currently selling api/chat access at a loss, but I see no reason to believe this.

Because infinite grow does not exist except in the deceptive narrative of neoliberal economics?

Everyone is SO focused on the immediate byproduct of this new technology, and there's so little questioning if applying it to business can actually make sense in the mid to long term. There are regulation pitfalls waiting to happen. We have no idea how web search and content consumption will adapt to this avalanche of cheap irrelevant content. Yet we give for granted this is the future of everything, based on the fascination with what these machines can do right now.

I'm saying this as someone who's also extremely fascinated and interested in the technology. This is the most excited I've been about tech in quite some time, but as I said, from a business perspective I've never seen so many business minds dive lemmings-style into an alpha-state technology before, and it's worrying how FOMO is pushing everyone to just take GPTs for granted!

You mean because in future it will be impossible to obtain a large high quality dataset due to AI contamination? Worst case you pre-train a model on pre 2022 data and finetune on a small, curated dataset of recent events.

And there are some important breakthroughs that have happened since GPT-4 started training, so it will almost certainly be possible to make a significantly better model with the same dataset and inference cost.

The LLMs are useful tools in their current state, right now. They produce text and images, which you can use freely after generation.

Nothing about their success is predicated on assumptions about the future. There's no risk associated with using an LLM today.

I mean at that point why not write it themselves? That's essentially what it boils down to, that they believe they understand what they want more than experienced writers and are willing to use a generative AI to try and work for them.

Of course the problem is that since they don't have the same editing skills or awareness, the moment they use an inappropriate line from the bot suddenly the reason why editors, writers etc exist becomes apparent.

Over the past 30 years I have worked many jobs in which writing was a central element, from tutoring at a community college writing center, to technical editing, translation, ghostwriting, science communications, and policy research. I haven't even used GPT4 yet, but just from playing with GPT3 it was very clear that the writing (sorry!) is on the wall. Yes, the prose can be stilted nd sometimes it just makes stuff up. But human writers, even good ones, are fallible, while also being slow and expensive to boot.

Even before AI, the internet experience had begun to undermine the traditional value assigned to good writing; it turned out that there are plenty of good writers who would publish their work for free. LLMs are already producing better content than most undergrads in their own fields, and can apparently do the same in every field, in multiple languages, as well as write passable code, make recipes, dungeonmaster RPG campaigns, spew out marketing bumpf, write surprisingly good creative mashups ("John Donne smoking salvia and rhapsodizing while clipping his toenails," or whatever), etc., etc.

Even if the technology doesn't improve significantly, I think this is going to have a radical impact on how writing ability is perceived - from a useful, even admirable, marketable skill to a novelty hobby, like chess. No doubt some very good writing will continue to be produced, and I'm sure LLMs will enable new forms of creativity in the hands of capable users. But the market for writing has always involved lots of jobs that only demand content that is just good enough, and if we're being honest, just crap really. Now anyone can have that for a $20/month subscription, plus maybe some light editorial touchup. The most talented and most specialized writers will still find work, but I guess their market value will be undermined by the combined effects of 1) the re-valuation of average-or-worse writing to zero, and 2) increased competition for paying jobs by a growing pool of under-employed/precarious, lean and hungry scribes.

Then you've probably been hiring people on the cheap, expecting low-quality work and getting it. The material you're getting out of ChatGPT is probably still very bad, but it is at least grammatical, and free.

If you don't mind that your content is meh, then you might as well use ChatGPT. I don't know who would want to read it, but apparently people do.

People would probably rather read insightful, interesting writing -- but not enough to pay much for it.

This comment was generated using ChatGPT 4 ^^

Nah GPT-4 wouldn't surround a double hyphen or emdash with spaces

But double hyphen is emulating endash, not emdash. So surrounding spaces are ok.

https://m.youtube.com/watch?v=yY1KdQNodN0

That would be 2 endashes, not an emulation of an endash, so surrounding spaces are not ok.

Although, I've always sucked at English so I'm open to being wrong.

Not sure I understand what you mean.

The parent comment to you said “writing -- but”. The “-” used there is different from the endash; “–”. And so I posit that “--” (double hyphen) is emulating “–” (endash).

ChatGPT4 uses (e.g.) Chicago style instead of (e.g.) AP for em-dash?

>Then you've probably been hiring people on the cheap, expecting low-quality work and getting it

If it works for their business needs, they could not care less, though...

If people cared for quality they'd be reading Tolstoy, Hemingway and the like, not listicles and Lee Child.

> If people cared for quality they'd be reading Tolstoy, Hemingway and the like, not listicles and Lee Child

All of these works are in wide circulation, with thousands or perhaps millions of readers each year. But which works do you think will have the most readers 100 years from now?

They replaced Upwork with ChatGPT. That probably is a step up.

If everything you do for money goes in and out over a wire, you're in trouble.

GIGO (garbage in, garbage out): it depends of your conversation [1].

[1] https://www.linkedin.com/feed/update/urn:li:ugcPost:70270343...

A non-insignificant amount of freelance writers are (were?) just writing spam blogs for SEO (aka “marketing copy”). Which in its turn funnels the search results into their actual money making business. Very rarely it’s done for the purpose of “good reads”. Writers who focus on “insightful reads” are more of columnists, substack people, personal bloggers whose shtick is writing themselves, rather than finding a freelancer.

In my opinion, for the former, the business is dead. Unless search engines figure out a way to derank GPT generated content (most likely will never happen) or they completely ignore random copy while ranking. For the latter, it might be exciting times, since people (I guess, me?) might start searching for actually human generated content.

When it is the race to the bottom, AI LLM services, like ChatGPT, Claude (Anthropic), Cohere.ai, etc are winning the race.

Open source LLMs are already at the finish line.

I found your comment very amusing and I didn't fully understand your comment. Can you expand on your reasoning?

What is the finish line? Why are contenders like ChatGPT or Claude "winning the race to the bottom" in your opinion?

Thanks in advance

> What is the finish line?

$0 / free is the finish line. Open source LLMs being released for free are undercutting the AI LLM services.

> Why are contenders like ChatGPT or Claude "winning the race to the bottom"

They're automating some humans in several industries and cheapening other professions. Why pay translators, copywriters, etc when a AI LLM can do it 95% cheaper or a locally installed LLM can do it for free.

The pricing of these AI LLM services are close to free and they can't cut prices any further. Once locally installed LLMs eventually equal the performance of that of GPT-4, Claude, etc the only logical step for those services is to slowly increase their prices and add more features as running these AI LLM services is very expensive.

Open source LLMs are essentially at the finish line of the race to the bottom but are admittedly not good yet, but will eventually be. Cloud LLMs like GPT-4 and Claude right now are good enough, close to free and are 95% cheaper than hiring a human journalists, copywriter, translator, etc.

Thank you.

Can I ask you a dumb question: Are you prompting it at all?

The stunted robotic writing is what you get if you don't prompt it. You can tell it to write like Hunter S. Thompson and Chuck Palahniuk for example:

As your cursor hovers over the downvote button, your heart races with anticipation. You can feel the sweat beads forming on your forehead as you prepare to unleash your disapproval upon the unsuspecting post. The adrenaline courses through your veins like a river of fire, urging you to click that button and unleash your fury upon the unworthy.

The moment you click that button, everything else fades away into a blur. The world around you dissolves into a haze of red-hot anger and righteous indignation. You can feel the power surging through you, as if you've become some kind of digital god, smiting the unworthy with your divine judgment.

And then, just like that, it's over. The rush subsides, and you're left with a sense of satisfaction and smug superiority. You've done your part to uphold the standards of Hacker News, and you've done it with all the ferocity and passion of a true believer.

In the end, it's not about the downvote itself. It's about the power it represents, the ability to shape the discourse and wield influence in the digital realm. And for those brief, glorious moments, you are the master of your own destiny, a force to be reckoned with in the online world.

I won't paste the output here (I've been downvoted enough already) but I will say that you inspired me to change one of my prompts from the other thread, like this:

Please write me a paragraph, in the style of Malcolm Gladwell, explaining why content written by ChatGPT is likely to be of higher quality than that written by a human writer.

The output is rather amusing.

This one is rather amusing as well:

Please write me a paragraph, in the style of Robert X. Cringely, explaining why content written by ChatGPT is likely to be of higher quality than that written by a human writer.

Aight, so cheap labor can lead to some cringe content, but ChatGPT? It's free and lowkey does a solid job. Not gonna lie, it's not a literary genius, but AI's leveling up with each tech upgrade.

Sure, ChatGPT's not as lit as those creative human writers, but it's a game changer for peeps on a budget. And remember, this AI's still evolving, so the sky's the limit!

Human writers got that unique vibe, no cap. But for some folks, ChatGPT is the real deal when cash is tight. So, don't throw shade at the AI too soon, it's out here doing its best.

Was the prompt "defend the use of chatgpt to replace human writers, parodying the way teens post on the Internet"?

I asked it to reply with some stank, then corrected it to only use slang from the last decade.

This comment reads like you told chatgpt to defend itself while talking like a zoomer. Especially when compared to your comment history.

Yeah, at first I just asked it to respond to the comment, but to put a little stank on it.

It was using some outdated terms so I then asked it to only use slang from the last decade and this is what we got.

This is stank level 7/10. You can turn the stank up or down.

I learnt the use of stinkyness as slang for slang today? I think

The more you know!

> THE AFRICAN AMERICAN VERNACULAR ENGLISH (AAVE) MEANING OF "STANK" "Stank" is an African American Vernacular English adjective that means "really stinks". “Stank” can also be an African American Vernacular English(AAVE) noun. One AAVE meaning for that noun is “someone or something who (that) really stinks”. No past tense is meant or implied by the AAVE word “stank”.

> Although the word “stank” in the saying “Put some tank on it” originally meant to have an odorous (stinky) smell, the contemporary meaning of “stank” in that saying means to add your own unique flavor (style) to something that you do and in doing so, make it funky (with "funky" meaning "very good".

> The phrase "add some soul to it" and "add your own flavor to the mix" are synonyms for "Put some stank on it".

http://pancocojams.blogspot.com/2020/11/what-african-america...

It's funny because I didn't google but assumed just that, I guess I hallucinated as finely as best current LLM's do

I can’t tell if this was written by ChatGPT or not (I’m guessing it was) and the same goes for a mounting percentage of comments over time. I suppose this means the enshittification has well and truly begun.

The GP would be better if they included their ChatGPT prompt.

This feels like a caricature of a caricature of a Xerox of a vapid MTV real world valley girl meets zoomer abomination. Absolutely brutal to read.

I think you underestimate how much corporate work boils down to string sentences together in a coherent manner. In the US, most college graduates get through school by writing essays and taking tests.

If a computer can do all of the essay writing and test taking… how many jobs did it just eliminate?

> how many jobs did it just eliminate?

Perhaps all, perhaps none. Tests (e.g. in school) are supposed to filter people. So a good test score in a well-designed test "correlates strongly with the human students' future performance" or perhaps "correlates strongly with the amount of work they tend to put in preparing". That's almost completely orthogonal to the test being about or requiring relevant skills to pass.

To illustrate this, very great chess players tend to be reasonably smart (e.g. you might expect them to be able to adapt and learn how to decently run a company). An AI that can wipe the floor with them can do no such thing.

Maybe this fundamental difference will go away with "general" AI and we may then judge it using human tests, but that's not a given.

>I think you underestimate how much corporate work boils down to string sentences together in a coherent manner.

I think you might be overestimating how many of those bullshit corporate jobs "need" to exist in the first place.

A consultant whose job is to rubber stamp an executive's decision may be able to use ChatGPT to write their justifications and keep everybody happy but if the exec went straight to the source they would completely undermine themselves.

This. You won't see New York Times articles written by ChatGPT, because the articles are condensed from a mountain of data. However, news articles that take a tiny fact ("celebrity x and y are dating again") and extrapolates 3 paragraphs of fluff from it, yeah, if you're writing those, your job is over.

Good point. Once in a while, I want to know who a celebrity is dating, or where they currently live. So I ask Google. The sheer number of celebrity news websites that are lower than ChatGPT 3 (!) boggles the mind. Utter drivel on these websites, but they must be making some money (adverts), else they could not pay for bandwidth / hosting costs.

Most of this content is going to a blog or newspaper that no one reads.

Scary as the prospect of job losses sounds, I feel like the real threat is that AI copywriting is just going to destroy content marketing altogether as businesses try to outdo each other with more noise and people begin to tune-out of the internet once they realize just how much of it is made by bots.

I think we overestimate how much normal people are plugged into the internet and I don’t think we give enough credit to the idea that trends in internet use might reverse course once they realize every person they talk to is a bot and everything they encounter was made by bots.

This, 100%! Also worrying: the complete disqualification of originality and human ingenuity in creative endeavors. While human creatives do follow trends, there is a constant strive for originality that pushes the bar forward. Now, in a corporate capitalist world where profit optimization is the prime goal above all else, we have the perfect optimization machine. The fact is that these machines are mixers. They never come up with anything original in the slightest. So if you remove the creation of original human content from which they fed initially what is left? All I can see is a destructive spiral of less and leas qualitative training that will inevitably lower the models abilities over time. In other words, these AI beating humans at their own game is a self-defeating effort in the long run.

The thing that gets me with so much of the ChatGPT writing is how often it does things like:

"In this essay I will..."

"First... Second... Third... Finally..."

"In summary..."

I have knee-jerk aversions when I see those phrases in any kind of writing. Before ChatGPT, those phrases would automatically make me think less of whatever I was reading. Now, it makes me think less of it and makes me wonder if the writer used a LLM to generate it.

In any case, while I agree that these models will have some impact on the Fiverr space and the exceedingly bland PR releases and SEO spam, the central problem of prompting is still applicable: in order to get quality results, you need to know what to ask for. In order to know what to ask for, you need to know how to evaluate the quality of what you get out. And, if you can do that reliably, you can probably generate it yourself without the LLM.

All that is to say, I see use cases as productivity tools and think the emerging field of prompt engineering will become more significant over time. I don't see it replacing a ton of non-Fiverr like jobs anytime soon.

It’s the Jane Schaffer model. I pointed that out a while ago - ChatGPT really likes to use it.

Before ChatGPT came out, I used to hire 2-3 freelancers every month for different writing projects. However, I recently realized that I haven't contracted any writers since then, without even consciously making the decision. Whenever I need some content, I can rely on ChatGPT to produce decent material. After some editing and refinement of prompts, it ends up being pretty excellent work. Moreover, using ChatGPT saves me time that I would have spent going back and forth to finalize a contract with a freelancer.

This is good? Humans wasting their time writing junk to fool the Google algorithm is a waste of their time. The problem is 1c goes to Microsoft instead of $10 to someone needing an income, so we concentrate wealth even more.

It isn't all junk though, it's stuff people spend their time reading. People like to listen to CHAT-GPT. I hear it all the time. I myself am not too big on talking to ChatGPT but I find that when I go to ChatGPT to experiment with it I tend to spend extra time just chatting.

Yes Chat GPT as a personal assistant is great. But tell it to create content spam and use that so someone else finds it when googling how to train their dog is less useful for the person searching.

Chat GPT doesn't waste words to keep you scrolling on the page like spam bloggers do. Chat GPT content is superior in pretty much every way.

Nor does it pad out writing because the publisher wants a certain number of pages.

It can waste words if your prompt it to.

“Write a spag bol recipe with a very long nanna backstory”

Then keep asking follow up questions for more text.

And it might keep doing it in the same window if you don't tell it to stop.

A lot of it is junk I don’t want to spend time reading but nonetheless have to look at in order to get to the content I want. For instance, try finding recipes on the internet. While some recipes are accompanied by real human stories and interesting metadata about the recipe, most are surrounded by junk text best stripped away. Similarly, consider the average news article: core factual content surrounded by junk text (often needlessly inflammatory to one group or another). For me personally, these and many other categories of information are best left to a program/AI that will omit the junk text when delivering me the information.

Most people do jobs that produce junk. Automating them all into unemployment would be disastrous.

nah there's still plenty of junk for them to do

That's a bit handwavey. And the problem for the economy (and the people that make up that economy) lies in the time in between them finding those other jobs.

Perhaps we should examine the culture we have that require people to do jobs that produce 'junk'.

It's not the culture, it's the people holding capital. Change the culture all you want, you still gotta eat.

You mean capitalism?

Only ChatGPT will have to read all the junk Google search results in the future when people go directly to ChatGPT anyways.

Next realisation: publishing ChatGPT generated content became useless because people use ChatGPT instead of search.

I saw this earlier where somebody used chatgpt and bard to stage a "debate". The responses were relatively predictable "high school essay" writing, which I can picture being better than some alternatives, if that's the regime you're interested in. As others have said, it's just a replacement for low-brow content that's not there to be read but exists for SEO or marketing. Cost rules, as long as it's passable english, as it's never going near an actual thinking brain.

Edit: forgot the link https://news.ycombinator.com/item?id=35659916

tbh, ChatGPT is as good as the typical SEO filler crap that’s littering the web.

Funny, I saw a post from one of those really infamous tabloids seemingly decrying how AI generated material was being used to spread lies about some celebrity’s health.

Sounds like they’re particularly afraid of being obsoleted.

People doing this don't realise that they're putting all their eggs into the end of the creativity basket. There is no way for AI to improve on itself like humans do, raising the bar over lifetimes and generations. While AI will certainly improve, I think you'll see it end up like midjourney or other stable diffusion systems which stagnate because they reach the limit of their creative learning data set. You need humans to keep producing novel content to allow AI to continue to improve. Once AI content becomes predominate and replaces human content I think we will start to see a plateau of "interestingness".

Stable diffusion stagnating? Compare the last models on civitai with the original model, it's night and day, and the tooling keeps improving too. There are still issues to iron out on both but there has been a lot of progress.

> There are still issues to iron out on both but there has been a lot of progress.

the progress has been so fast that i cannot foresee any problems/issues that wouldn't be fixed within a few years.

The stable diffusion finger mangled issue is almost fixed already (not quite yet, but very nearly there). And with controlNet, posing consistency is also something that is easily fixed.

LLM will have a similar cadence.

What? Multi Controlnet with high weight gets perfect fingers every single time

Midjourney also has just gone from horror blobs to photo realism. I really don’t know how anyone could say they’re stagnating

AI can improve using the same corpus, there are many examples of that. Humans will still curate content as well.

People have noted that at the extrema AI will cause problems for copyrights and people won't publish original content because it will be consumed by these AI's. All of which might be true. But it did make me wonder if AI might cause a big shift in our political/economic environment.

For example, could Ai be considered a public work and a model where a key aspect of government is to reward contribution to it. That is, people would focus on generating content for AI and be paid by the government to do so with some governing body that determines the value of contribution. The idea being that enhancing the AI is a public good. At least until AI can generate its own novel and useful content. At which point things probably takeoff in a way that looks like sci-fi.

Freelancers like this are hired to produce spam content that is basically keyword stuffing. They are not hired to create a literary masterpiece. In that context "never hired a writer better than ChatGPT" is not surprising at all.

Lately I’ve been practicing a new form of exercise to me (mace and heavy club swinging, if you must know). While most of my learning has come from online videos, I’ve also bought a few self-published books by people who made the videos. The books are helpful, too, but they are not written well at all—not surprisingly, since the authors are professional fitness trainers, not writers.

A year ago, I would have wished that they had hired a good editor to clean up and tighten their prose. Now I would just recommend that they run it through GPT-4. That would make their books much more pleasant to read and learn from, and it would cost them a lot less money.

"Georgia Austin, who made $2 million in two years copywriting for brands..." wait what, that doesn't seem realistic?

No, it's a cherry-picked example of 0.0001% of writers. Otherwise there would be no article to write.

Related professions like scriptwriters and script doctors are the same. Most people will make very modest amounts. But a handful can create a huge amount of value with just the right structure and words-and assuming they have the right connections.

Has anyone else noticed news articles taking another steep dive in writing quality recently?

The first major dip was right in early lockdown era. Now i'm finding articles to have increasingly less relevance/substance related to their headlines in the period of the last month or so.

It seems like the easy answer is AI, but I think AI would do a better job. Am I wrong?

No, I haven't noticed this in the media I read. Perhaps you could share some examples of what you mean?

ChatGPT has turned me into a better writer. I am pretty good at converting information but it is often a dry writing style. ChatGPT can help make the style more engaging while still conveying the same information.

The world is starting to realize most people's job are not that complicated, including high standing positions, and that on top of this, a huge number of workers are not that great at their job.

Well, yes, the world runs on average. Its runs on meh.

AI will be quite good at meh.

Why do people keeping saying AI tools are free? You hire a freelancer for their time.

Doing it yourself has a time opportunity cost.

Hiring a freelancer also costs time. If using an AI tool is faster than the hiring, you already won.

I’d rather negotiate with and edit ChatGPT than negotiate with and edit a freelancer.

You still need to brief the copywriter on what you want. Is that any less work?

One day we're going to have a blog post reach the front-page of HN which was written with the help of by GPT-4, if it hasn't happened already. And then it will happen again and again...

I wonder if we'll even see a post almost entirely written by GPT-4. I don't think that will happen (except in a "look what GPT can do!" way), but time will tell :)

This reminds me of this xkcd : https://xkcd.com/810/

Me: Please write me a paragraph explaining why content written by ChatGPT is likely to be of lower quality than that written by a human writer.

ChatGPT: As an AI language model, ChatGPT generates text based on statistical patterns it has learned from large amounts of human-written text. While it has the ability to produce coherent and grammatically correct sentences, it lacks the human touch of intuition, creativity, and emotions that make writing engaging and memorable. Additionally, ChatGPT may produce inaccurate or biased information if the data it has been trained on is flawed or incomplete. It also does not have the ability to understand the nuances of human language, such as sarcasm or humor, which can lead to misunderstandings. Overall, while ChatGPT can be a useful tool for generating text quickly and efficiently, it is unlikely to match the quality and depth of content produced by a human writer.

Well, at least it's honest about that. shrug

To be fair though, let's give it a chance to plead it's case the other way:

Me: Please write me a paragraph explaining why content written by ChatGPT is likely to be of higher quality than that written by a human writer.

ChatGPT: As an AI language model, ChatGPT has the ability to generate large amounts of high-quality content quickly and efficiently, without the limitations of fatigue, bias, or emotional interference that can affect human writers. With its vast knowledge base, ChatGPT can access information and synthesize it in ways that would be difficult or time-consuming for a human writer to do. Additionally, ChatGPT's writing is free of typos and grammatical errors, and it can produce text in multiple languages, making it a versatile tool for content creation. ChatGPT's writing is also free of human biases that can affect human writers, making it a useful tool for producing objective and unbiased content. Overall, while ChatGPT may lack the human touch of creativity and intuition, its ability to generate high-quality content quickly and objectively makes it a valuable tool for content creation in many contexts.

TBH, 99% of human writers would struggle to produce this analysis at all, let alone efficiently.

This is why so many people are turning to ChatGPT.

At least tell us if this is GTP4 (the paid version) or not. I know GTP can say amazing and relevant things, but at this point quoting it is like copy pasting large sections of Wikipedia without added insight. We can all go talk to GTP ourselves if we want.

No, that was GPT 3.5. But here's the GPT 4 version, if anyone is interested:

Me: Please write me a paragraph explaining why content written by ChatGPT is likely to be of lower quality than that written by a human writer.

GPT 4: While ChatGPT, based on the GPT-4 architecture, is an impressive AI language model that can generate human-like text, it is important to acknowledge its limitations in comparison to a human writer. First, ChatGPT lacks the nuanced understanding of context that humans possess, which can lead to inconsistencies or inaccuracies in its responses. Second, its training data, which is only current up to September 2021, may leave it unable to address recent developments or ideas. Third, the AI model can be sensitive to input phrasing, meaning that slight changes in the user's prompt can yield vastly different results. Finally, it cannot draw on personal experiences, emotions, or empathy in the same way that a human writer can, which may make its content less engaging, relatable, or authentic. While ChatGPT can produce impressive text, it is essential to recognize the areas where it falls short of a skilled human writer.

And the other way around:

Me: Please write me a paragraph explaining why content written by ChatGPT is likely to be of higher quality than that written by a human writer.

GPT 4: Content generated by ChatGPT, utilizing the advanced GPT-4 architecture, has the potential to surpass human writers in certain aspects. Firstly, its vast knowledge base, which encompasses a wide array of subjects, allows it to provide well-informed and accurate content on diverse topics, often exceeding the expertise of a single human writer. Secondly, its ability to process and analyze large amounts of data quickly enables it to deliver high-quality content efficiently, often in a matter of seconds. Additionally, ChatGPT's impartial nature can reduce the risk of personal bias or emotions affecting the content, promoting a more objective and factual output. Furthermore, the AI's consistency in grammar, spelling, and writing style ensures that the generated content maintains a high standard of quality, free from common human errors. While it may not replace the creativity and empathy of human writers, ChatGPT offers distinct advantages in specific contexts.

Doesn't appear to be a huge difference at first blush. YMMV.

“ without the limitations of…bias”

Human writers also have a hard time recognizing and acknowledging their own bias, so I guess this is pretty spot on.

Reply on news.ycombinator.com