Nah, it’ll be just a new, higher level of abstraction (and provided we even solve the hallucination problem, which is not a given). Engelbart’s and Parkinson’s laws at it once again.
In other words, the new work fill the same time, and we’ll just be doing 10x of it (if we can solve the current challenges with LLMs). The human mind is unsurprisingly adaptable to changing conditions and great at abstract thinking, which is what ironically distinguishes it from these machines.
We also tend to severely underestimate our own abilities and overestimate the importance of the tools.
> We also tend to severely underestimate our own abilities and overestimate the importance of the tools.
Even on HN, this is rampant. I wonder, is it because people are so used to interacting with most other humans solely through a chat interface that they perceive ChatGPT as being close to a real AGI, almost indistinguishable from human intelligence?
Pretty much :-) Text generation (and rudimental comprehension) is perceived to be a sign of higher intelligence, simply because it's our natural human UI.
Ironically, we tend to limit our imagination on what _we_ could do that computers won't, while simultaneously overuse it to imagine what they _could_ do but won't.
One very good example of this is humor - LLM models at present are terrible at generating anything but dad jokes, and are even worse at their attempts to explain "deep" humor. Gary Marcus had a recent episode on it, very interesting insights from top experts in the field: https://podcasts.apple.com/us/podcast/can-ai-make-you-laugh/.... They make a case that LLMs were specifically optimized for text generation and not explanation, while humans are almost the opposite.
I believe somewhere in there is also our tendency for confirmation bias and overlooking imperfections where we don't expect some. AI may generate nonsense details, but when we look at the overall text, we don't notice them since we're just awed by the output, and think if it looks flawless it must be flawless. Same thing as with mentalist tricks.
And finally, we only think of our intelligence as thoughts carefully animated into sentences, while there's this whole "iceberg" level of feelings and unconscious cognition, which we don't even notice ourselves, but it would be extremely hard to reproduce in AI and we have not even the slightest idea how.
I worked in a semi automated assembly line. We,as workers were pretty good at keeping the robots running and knowing their quirks and my supervisor was even more knowledgeable. Despite this we still had to call up an engineer daily to trouble shoot why the robots were revolting. AI is not remotely near solving this.
Disagree - AI is getting much closer to solving this fast.
Latest generation robotic arms with vision are taking on a greater variety of tasks (e.g. bin/case picking from a mixed tote), more reliably, with a better ability to adaptively recover failed tasks.
The progress here are about not needing parts/items to be in a predictable/consistent location and still understanding how to handle/retrieve them and perform actions on them. This stuff is in sites right now.
picking stuff from bins is a far stretch from a full assembly line production. Latest generation arms are surely being developed but the are far from fitting affordability profiles for the average American factory. It sounds great when you are reading it but fabrication is not like picking stuff from bins. The colossal amount of programming for every human task that is performed in a factory that builds compressors would be a dauntingly expensive enterprise even without counting the physical robotic equipement that would be needed. Theory and practice don't always correlate.
Heres a small example. In one position the compressor gets scanned and rolls up to the operator.The label needs to be smoothed around the rounded compressor (it has been left standing out at previous stations because the scanners have a hard time catching them otherwise). The scan reads the style of compressor and this lights up parts bins so the operator knows which parts are required. There are two different types of screws and also a ground screw so the robot would need to be fitted with the ability to pick scews and screw them in. They would also need to apply plugs clips inserts, caps, connect wire ends, apply two different kinds of stickers and put a clip on cover that requires finesse to get clipped in all the way without cracking the plastic. Also the the final piece needs to be checked for paint flaws or visual weld defects. All these actions are done by a good operator in under a half a minute. And this is just one station in one part of the assembly line. Sure its possible, but practical or cost effective, not really.
Whenever I hear about AI taking knowledge work, I always think of that show Dirty Jobs. Jobs like 'sewer cleaner' are not going to be automated anytime soon. So eventually all the fun jobs will be AI driven and humans will be left cleaning shit off the walls.
“Think of AI as generally acting as a high-end intern,” he said. “Jobs that are mostly designed as entry-level jobs to break you into a field where you do something kind of useful, but it’s also sort of a steppingstone to the next level — those are the kinds of jobs under threat.”
Every bit of this article worries me. Not because I think it's coming for my job, but because it's coming for so many other jobs.
"Experts say that even advanced AI doesn’t match the writing skills of a human: It lacks personal voice and style, and it often churns out wrong, nonsensical or biased answers. But for many companies, the cost-cutting is worth a drop in quality."
That's the part that makes me think this is going to cause some serious job losses. "the cost-cutting is worth a drop in quality." All of the AIs are pretty ok right now, and we're already seeing some (probably not many?) jobs cut because the stuff they make is "good enough". I'm worried what kids just starting college now will face in 4-5 years when they're trying to start a career.
I think a lot of the jobs that go will be the ones that popped up recently. The low level white collar ones. Social media content writers for example. That tends to be a very well paid job compared with the skill necessary to do it and the benefits it provides. I imagine most of these jobs will just get replaced with something else that the AI can’t do and that only new grads will accept doing.
For your kids, I think you’re imagining more progress with AI than we’ll have in the next five years. I think the changes of the last 3 years will be much more impactful on new grads. I recently switched role and learning remotely is a lot more difficult for various reasons. I’m already well into my career so I imagine for new grads it will be incredibly difficult. I’d worry more about that than AI provided they’re going to college to study something “useful”.
Couldn’t adversarial networks be used to continue to improve quality? I guess we could wind up with a bunch of humans in the loop to help guide learning improvements too, seeing as we’ll all be unemployed (sort of joking).
There’s only so much information content that exists on the Internet. There’s a limit to it, and one cannot simply produce more info once that limit is reached, it goes against the entropy so to speak. And then, “garbage in - garbage out”.
In addition, I think we’re living in a huge bubble and are sort of entitled enough to think that Internet = human knowledge. So many jobs and skills are simply not transferable even using human language (and even if they hire an army if “trainers”). This is one of the core problems with the current incarnation of “AI” - tacit knowledge has always been the stumbling block, even back in the expert systems days when everyone was thinking what we’re thinking will happen today.
So until the machines can somehow gain consciousness, autonomous and nimble movement, the will to live and all of the deep levels of the human mind that are outside language, I think we’ll be pretty safe :-) And for now, on all of these parameters they aren’t even close to a cat.
Exactly, AI can regurgitate all the beautiful sunsets it has been trained on. An artist experiences a sunset and is inspired to create things inspired by the feeling of a sunset. These things could be textiles, furniture, abstract art, dessert, any number of functional items the color of the sunset, a short story, a mimi series....The misconception is that creativity is created by studying and copying past creativity. Creativity comes from living experience and is applied thoughout all aspects of our lives. The belief that AI trains the same way humans do is a sad misconception of the human experience.
Right? I think it's just the Silicon Valley again and again trying to push its single-minded vision that computers should be used to "solve" all of our problems. They see everything as a nail with a "problem" to solve with their hammer, and completely disregard that there's life outside their bubble in the SV.
> I'm worried what kids just starting college now will face in 4-5 years when they're trying to start a career.
Why? Traditional family unit is over and generations are at war. Let them figure it out. Meanwhile let's not have kids without a clear plan for their future.
If AI is the intern, how do the economics of entry level positions work? And if they don’t, how do people enter the workforce at all? More schooling? Only hire PhDs?
Well, we have to work with what we have :-) It would be highly unlikely that something new violates the established trends over thousands of years.
It's not surprising that the human race has survived for this long - we always find ways to adapt, no matter what challenges history throws at us. This is what distinguishes us (and I suspect will always do) from the machines.
> It would be highly unlikely that something new violates the established trends over thousands of years.
What now? Established trends have been regularly violated over thousands of years.
> It's not surprising that the human race has survived for this long - we always find ways to adapt, no matter what challenges history throws at us. This is what distinguishes us (and I suspect will always do) from the machines.
You know what would count as an adaptation? Large fractions of population becoming immiserated and either dying off or living in poverty at the margins.
I think you're severely overestimating the abilities of this "AI" and underestimating the abilities of the human mind here :-)
However, if you'd like to be pessimistic on it I won't argue, I have that tendency too ;)
And to lighten up the mood, I highly recommend listening to this episode with Gary Marcus, where he invites experts in the field of humor to discuss how distinctly bad the LLMs are at explaining what comes naturally to us: https://podcasts.apple.com/us/podcast/can-ai-make-you-laugh/...
> I think you're severely overestimating the abilities of this "AI" and underestimating the abilities of the human mind here :-)
"This AI" seems perfectly capable of resulting of extending the kind of economic immiseration that has already affected many blue collar jobs into many types of white collar work (ironically enough, the kind of work those blue collar workers were told to pick up).
Past this iteration of AI, who knows if the advances will peter out, but it's foolish to assume things will be fine because other things were adapted to in the past. That's kinda like assuming you won't die before having kids because none of your ancestors did.
I won't engage any further. I see that I won't be able to convince you in the opposite, so there's no reason for me to try to "defend" my argument :-) And your "command" is kinda rude, regardless.
> I won't engage any further. I see that I won't be able to convince you in the opposite, so there's no reason for me to try to "defend" my argument :-) And your "command" is kinda rude, regardless.
You mainly been making vague, unsupported assertions backed by emoticons. It's a stretch to call that an "argument," and arguable you haven't been really engaging from the start.
I don't think you can convince me because I suspect you don't really have any support or specifics for your position to provide.
> As a matter of fact I do, if you read my comments in other threads, before jumping to conclusions without knowing me.
Come on. If you have them formulated, you should have responded with them. It's not that hard.
> Regardless, I flagged your comment since it really doesn’t follow HN etiquette on respectful arguments.
Again, come on. The disrespectful thing is expecting someone to go on a scavenger hunt just to find out what you, some internet rando, should be saying in your half of a conversation thread.
It's also just bizarre for you to note you flagged my comment. Am I supposed feel intimidated or ashamed or something?
> Established trends have been regularly violated over thousands of years.
I don’t think that’s true. The archetypal constructs of human cognition have been unchanged. Hence the enduring relevance of myths. It’s always the same stories playing out.
>> Established trends have been regularly violated over thousands of years.
> I don’t think that’s true. The archetypal constructs of human cognition have been unchanged. Hence the enduring relevance of myths. It’s always the same stories playing out.
I think you misunderstand. I didn't say everything has changed, just that many "established trends" have been.
I mean, at some point it was a well established trend that all humans hunted and gathered for their substance, then that trend was violated.
At some point, the trend of "we will find something productive to do with all these people who's jobs have been made obsolete" may be violated. And it seems like that's getting more likely the more rarefied the remaining valuable skills become.
> You know what would count as an adaptation? Large fractions of population becoming immiserated and either dying off or living in poverty at the margins.
> Mere survival is the wrong metric.
People don't have to live in cities. Knowledge work has brought many people to cities and the absence of it will reverse that trend. It is welfare that keeps impoverished people living in cities.
> If AI is the intern, how do the economics of entry level positions work? And if they don’t, how do people enter the workforce at all? More schooling? Only hire PhDs?
They don't, and in 10-20 years corporate leaders will bemoan the shortage of high-skilled experienced people that they created through their own decisions. Of course, they won't take any responsibility or change their approach, though.
> They don't, and in 10-20 years corporate leaders will bemoan the shortage of high-skilled experienced people that they created through their own decisions. Of course, they won't take any responsibility or change their approach, though.
It's a free rider/prisoner's dilemma problem.
If everyone cooperates then we all come out ahead. If you all cooperate while I defect then I end up even better (I don't have to train anyone). But if we all defect then we're fucked.
It used to be that there was entry level work that needed to be done by humans and so you didn't have this problem. If the entry level work can be done by AI then you need something else. Either government coordination and incentives to hire humans (this is the kind of things gov't is good at). Or people stay in education for longer/education becomes different.
Temporarily, maybe, until misguided decision makers realize they could not feasibly replace people with a half-baked ‘AI’. And I do think we (the community lurking on HN) are living in a huge bubble - there’re so many more jobs out there that we have not even a slightest idea how to replace.
On top of that, the biggest unaddressed problem that everyone seems to pounce over is tacit knowledge - it cannot simply be transferred even between humans, using words, gestures or any other media. It can only be gained through the human experience, by doing stuff. It’s the same old problem from the expert systems days. So we cannot hope to replace them with AI, at least not until AI somehow becomes self-aware and gains all of the non-surface depths of the human mind, and lives through the same experience.
Perhaps the free tier should be jettisoned (at least in the workplace/enterprise?) in an effort to curb this? Obviously if it’s free then some companies may make these sorts of decisions. But if they have to pay then at least they will think more about eliminating their “lone” writer..
The long-term play is probably to capture a large portion of the salary savings these companies get. But you need to get companies hooked and dependent upon your product first. Once lock-in is achieved, they will ratchet up prices as high as possible.
Marketing and advertising copy as a service. Automated A/B testing and learning based on integration into web analytics and e-commerce systems. Imagine if your advertisements, product descriptions, white papers, and corporate blog posts automatically learned from customer behavior and revised themselves to maximize customer engagement and revenue.
There aren't enough of those sorts of jobs to accommodate everyone, though. And how large could the market for that sort of thing be when a large percentage of the population can no longer earn enough money to pay for them?
Reflects the attitude of the company toward its workers.
I have one friend who, instead of firing her, her company asked her to use ChatGPT and increased her work load. Smart managers will augment their workers, unless, of course, the worker was poor-performing to begin with.
Increased work load is just as nightmarish to me. In any case as improvements continue that would probably just be kicking the can down the road and abusing workers, before firing them later.
Automation should let people work less. Not just give rich people leverage to work us harder. Making people juggle more tasks at once with the AI will just burn them out.
Fortunately or not, history has shown time and time again that Parkinson's law still holds - work will always fill the allotted time. Pretty much the only way to decrease it is to mandate fewer working hours, like is starting to be done in Europe, for example.
in other words they gave her an intern who is fast but of in ferior quality that she now has to direct and moderate on top of the good quality job that she already does so that she can increase her work load for the same pay. Sounds like fun.
US is kinda fucked in this regard already. The median income is what, 30k? Unless your product is cheap or a necessity of some kind, you've probably already lost half your potential audience.
And with that income it will be hard to support social programs like pensions. So the pyramid scheme might just collapse and retirees are screwed. In that sense the talk about UBI looks completely bonkers if no high earners left to fund it.
There are plenty of people in the world and the median income globally has risen rapidly over the last few decades.
All it means is that wasteful western lifestyles can't be continued while producing little of value.
They will collapse as fast as they have risen. I don't see how 3rd world economies will survive without the westerm middle class. I guess just slaughter each other as humanity always did with excess populace.
You will mortgage your degree or your work license in order to pay off your purchases, that have already been accumulating debt from the split payments for years.
The government will later offer debt relief to those who agree to be permanently sterilized. It will offset your debt by giving huge amounts of carbon credits for the emissions your children would have been guilty of.
> “In every previous automation threat, the automation was about automating the hard, dirty, repetitive jobs,” said Ethan Mollick, an associate professor at the University of Pennsylvania’s Wharton School of Business.
What an odd statement. Knowledge work has been automated for as long as I can remember. Compilers, automated QA, data entry, etc. In many fields, more labor demand resulted from productivity gains (software); in others, demand didn't rise to compensate (bookkeeping).
The question is where is additional demand going from higher savings given productivity boosts. I don't see more demand for HVAC technicians. Maybe dog walkers.
But all the examples you gave (compilers, automated QA, data entry) fall firmly within the "repetitive" marker. These were all pretty much rote jobs where a person was essentially asked to be like a computer and follow a simple algorithm.
But ChatGPT is now killing jobs that are solely about creativity, like copywriting. That is a new, and rather unexpected, turn of events. I saw a good quote, I think it was on Twitter: "What kind of world have we built where the humans slave away at physically demanding jobs for minimum wage, while the computers write poetry and paint all day."
Don't agree with that at all. If it were simply "repetitive and non-creative", why hire copywriters at all, why not just have the engineers or product managers write the copy?
I've worked with good copywriters, and was impressed with the quality of writing they did. The point is that ChatGPT excels at the type of creative writing that previously we thought was solely the domain of humans. I think you are getting things backwards if you think that because ChatGPT can do it that it is repetitive and non-creative.
"Knowledge" jobs have been automated a long time. Even before digital computers. Be it census or what. Then with computers insurance, banking and so on. This automation then moved to smaller and smaller companies.
Actually when you think about it, even clerical work was "automated" by printing press.
Automation of production of generic content is bit scary, but it might free up resources or just show what the truly value of that work was.
I thought this was a good and fair article on the topic, which is kinda rare because often you just get tons of speculation. This article was just filled with real people who actually did lose their job due to ChatGPT, and I think it just shows the tip of the iceberg.
I would really be worried right now if I were a marketing copywriter. I've seen (and posted about) examples where ChatGPT still falls behind human ingenuity, but the fact is that it really only falls behind some of the best of human ingenuity. Just guessing, but I'd believe that ChatGPT is as good as or better than 75% of copywriters, and even for that upper tier of copywriters, probably only about 10-15% of the writing they do is in an area where the additional quality makes a difference.
No one actually reads most of that copy anyway. It's the knowledge work equivalent of digging a hole just to fill it in again. Totally pointless, and done more out of inertia than any real business value.
Saw a comic in the early days of ChatGPT, I wish I'd saved it:
What the applicant writes: "I want this job."
How the application is sent: "Dear Sir or Madam, I am writing to you today to inquire about the position of blah blah blah..."
What the recruiter reads: "This guy wants this job."
I'd like a tool I could point at a modern tech company website to read through all the meaningless ad copy and buzzwords and spit out a two-sentence summary of what the thing actually does.
ChatGPT can already do this. In fact, I think it's much better at this kind of summarization task than it is at novel text generation. You'd have to write some glue code to automate it via the API, though I'd be surprised if there aren't open source tools to do it already.
As a demo, try this prompt; you might have to trim some to stay under the token limit:
Summarize the major arguments and sentiment of the following Hacker News posts. Provide your answer in a brief, bullet-point format:
Microsoft is busy integrating GPT into all of their products. I expect the next release of Outlook to have a button to generate an AI summary of long messages.
The "Statement on AI Risk" by Sam Altman and wealthy individuals fails to address the potential impact of AI on job security. Rather, it focuses on concerns regarding war and human extinction. Despite my attempts to raise awareness about the issue, I have received negative feedback. It's important to consider how the everyday struggles of families may affect their ability to purchase products and services such as phones, streaming, gaming, and others when AI takes away their white-collar jobs. Maybe there will be war because there are no jobs while rich folks are getting rich. I hope I remain wrong.
They also don't fail to realize when you ultimately drive 80% unemployment due to AI replacing everyone's job. That war is going to be internal and it's going to look like the French revolution.
It will only look like the French Revolution up until the point where fully-autonomous battalions churned out by fully-autonomous factories fed raw materials by fully-autonomous mining and transportation operations make the owners of AI untouchable.
There comes a point at which the plebs can't do anything about their fate.
But that's why we have TikTok, Instagram, Netflix, cheap Corn Syrup, and Porn. You keep the plebs hedonistically satisfied while you slowly consolidate your power and protect it with an autonomous army. THEN you pull the plug.
> The "Statement on AI Risk" by Sam Altman and wealthy individuals fails to address the potential impact of AI on job security.
I think that was very intentional. Put the focus on scary-sounding things that aren't actually an imminent risk in order to distract from those things that are.
I know someone who does copywriting for cybersecurity firms. She writes those seemingly mundane statements these companies release. I've always wondered if anyone actually reads them. She admits that most of her work involves taking a template and slotting in the new information. It's stuff like "Company X's response to security threat Y," which often doesn't contain anything particularly insightful. It usually just states that the company is aware of the threat and is working on it. Nothing technical or useful.
I'd be sad to see her lose her job, but at the same time, I can see how unfulfilling this job must be for her. She often feels that the work she does doesn't matter to anyone.
To me, solution is not to invent more bullshit jobs. We need a better system for humans to share the fruits of millennials of progress together.
This is wishful thinking at best. She's fending herself and she's making an honest living out of it. This system is not perfect but what is the alternative? There's no current solution for what lies ahead (potentially job loss on an scale never seen before).
I'm seeing government outright banning LLMs from certain industries just to avoid social unrest.
I wonder how this is going to affect the cost of accessing ChatGPT and other AI systems. If ChatGPT can replace a copywriter, who may have an annual cost of $30-$50k per year, then it could easily justify a cost $10k or more per year.
The current model seems to be just like Uber. Get into as many markets as you can. Provide the service below cost. Then jack up the price and make money.
As someone who uses ChatGPT as a learning tool for various programming tasks, I'll miss it if it becomes too expensive. Its a helpful tool, but not worth more than $20/month.
I think unlike Uber, the future floor for costs is probably very low. You can get insane improvements on inference costs/speed when you dedicate time and money towards optimization.
Uber has hard constraints that are difficult to significantly reduce with scale. GPT is already approaching adequate on consumer grade GPUs. Pretty sure the price is going to trend towards zero, no matter what the salary is of the employee it replaces.
How can you internet in {current year}, and seriously think we need more content writers? We’re not talking about top-tier investigative journalists blowing scandals open and speaking truth to power. We’re talking about the meaningless banter filling space between ads that turn a simple recipe search into a dystopian nightmare.
I hope and trust that they will find meaning in a career fixing and building and caring for physical things that people actually want to pay for. This is a step forward.
Couldn't agree more, and my work is largely content writing.
For content to have any value, it should contain unique information or an especially good and useful presentation of existing information. Otherwise making that content is just worthless busywork. Like the example in this thread of a poster making a press release. How could AI add any information that is of any value?
The same goes for AI generated images. If your images are not extremely relevant you are doing it wrong. Use images of your product, not random people smiling. Stock photos, Corporate Memphis art and AI images are a negative value for the information you're presenting.
Just as you, I hope companies some day understand that their worthless filler content is just getting in the way of their sales.
Thank you for that perspective. From the AI side, I think it’s funny how there can be so much simultaneous interest in both “AI writing”, which is exactly as you say, and “AI summarization”, which is like inverting the AI writing function to retrieve the prompt data.
An absurd, but seemingly likely, end state of this is an arms race between the two technologies, completely transparent to the user, and resulting in roughly the same user experience, which is always set to the limit of their tolerance.
This feels largely like Fear-mongering to me, and it can't be helped but noticed that it is coming from a source that is likewise threatened by automation. Fear of change and fear of the future sell newspapers/subscriptions/get people to quit.
I don't doubt the threat to copy-writing and similar jobs, but I also suspect that the call/response nature of the advertising industry is going to result in people being inoculated to ad-copy written by ChatGPT, and it will start to sound as hokey as old ads you find in computer magazines.
As someone who makes a good portion of my living as a writer, I choose to embrace ChatGPT to make me more productive. That's how I stay relevant.
For example, today it helped me write a press release. We had dialogue back and forth. I told it to ask me thoughtful questions, and it did. After 5 minutes of back-and-forth, I had the bones of a high-quality press release put together. I then edited it until it met my standards, and submitted it.
This process would have previously taken me a few hours. Today it took less than 30 minutes.
As a result, my value to my employer has gone up because of ChatGPT. I'm able to get a lot more done with it, and they'd be fools to simply replace me with the bot.
In other words, and I'm hardly the only one saying this: ChatGPT is not a replacement for people. It's an augmentation. It's a bicycle for the mind, to steal Steve Jobs's famous metaphor.
I think if you're a writer who refuses to embrace this new reality, you should absolutely be worried. But if you embrace the inevitable, the combination of a skilled copywriter + ChatGPT is incredible. It's my writing partner. I'm still in charge, calling all the shots. But my writing partner helps to prevent writer's block.
Often, I won't use a single word it produced. Instead, it gives me a starting point. Sometimes that starting point is little more than "I need to do better than what this machine produced, because this is crap." But it's soooo much easier to improve on garbage than it is to create out of the blue.
So, I don't think AI is coming for our jobs as writers. If used well, it's the best writing partner you could ask for.
> In other words, and I'm hardly the only one saying this: ChatGPT is not a replacement for people. It's an augmentation. It's a bicycle for the mind, to steal Steve Jobs's famous metaphor.
This is the analogy that jumped out to me as well. LLMs seem to be more of a vehicle than a destination. Like a bicycle, LLMs need to be operated with the direction and effort of an operator to ensure it stays on course and doesn’t veer into traffic.
Continuing the analogy, I suspect that LLMs will open up traditional white collar jobs to more people. More people can get to the destination on time with the aid of a bicycle than on foot.
If you can suddenly produce the content of three writers, two other writers are getting fired, and you'll be tasked with the work of five, under the same deadline. That's the way capitalism works.
Seems like an overly cynical take on capitalism, and I'm not exactly capitalism's biggest fan.
All businesses are in the business of maximizing profits. If they can be more profitable with fewer writers, of course they'll reduce staff. But if X content produces Y results, one should expect that 4X content will produce 4Y results. Profitable activities get invested in. Cost centers get optimized.
ChatGPT lowers the bar for content creation, which actually raises the bar for producing content that will stand out. If you're just copying/pasting content from ChatGPT, you'll have a lot more content, but it won't stand out. It's content for the sake of content, as opposed to content that's meant to engage and persuade a person to spend money on your products.
This is especially true with LLMs. They can't come up with truly creative content. LLMs, by design, return the most average content imaginable. This has tremendous value in the hands of a skilled writer/editor.
Remember: The purpose of marketing is to make your brand and your products stand out from the crowd. You're not going to achieve that by producing what is, by definition, the most average content possible.
ChatGPT makes me a better writer. That's all there is to it.
Businesses that don't believe content can sell shouldn't have any writers at all. Those that do should invest in AI-competent writers more than ever.
I maintain that AI is only a threat to me as a writer if either (a) I'm a hack who produces low-value content, or (b) my employer is stupid. I'd rather not work for stupid, anyway.
Exactly! And this applies not just to the writers, but any (and I mean Any) job out there - 10x-ing human output would just mean 10x more work could be done with the same amount if resources (including time and cynically, people), so a competitor won’t stay idle exploiting that. I like Parkinson’s law at play here.
Let's say it costs $100 for a piece of content, like a blog post. (I'm not suggesting that's the right price, I just like easy math.)
And let's say your return is $200 in sales, leaving you a profit of $100.
If I could double my money like that with every article, I would gladly pay $10,000 for 100 articles if the return is then $20,000. That's a money-printing machine.
Now let's say the return isn't $200 per article, but $105, for a profit of just $5. I wouldn't pay a single penny for it, because I have better ways to invest my money.
This applies to anything in business. Whether it's content, technology, inventory, people, real estate... whatever. If the return is great enough, smart business owners will want more of it.
It has nothing to do with how much time it took or how many people were employed.
It has everything to do with how much it cost versus what the expected return would be.
That's all business is. Turning $1 into $2.
Show me a process that does that, and I'll invest every dime I've got.
Now, let's go back to the original example: $100/article for $200 return. If ChatGPT allows my writer to produce 10X as much content at the same cost (which I doubt... it's a bicycle, not a jet plane, but I digress) without any loss of quality, I would anticipate my returns to be $2,000 for every $100 invested. That would be incredible. I wouldn't just invest every dime I have but I'd max out every dollar I could borrow as well.
BUT: That assumes each piece of content can be just as (or more) effective as those that came before. There's a limit to how much anyone's audience really wants to hear from them.
The answer to this is to invest in finding new audiences. New niches that would want your products. High-quality expert content is a fabulous way of attracting communities of like-minded people. So when you get to the saturation point with one community, there's an opportunity to find another, and then another, and then another, and so on and so forth.
It all hinges on whether the content is actually doing the work it's supposed to be doing. Things that make money get invested in. Things that cost money get optimized or eliminated.
If I was a writer and I lost my job to ChatGPT, I would take a long hard look in the mirror and ask myself why my employer didn't see my value anymore. Was I just producing content for the sake of content? Was my content profitable for them? Was the quality of my writing so mediocre that a machine did it just as well, if not better? If my work generated profits, why would my employer cut me out? Why wouldn't they want even bigger profits? Aren't they in the business of making money?
--
Now, I responded to your comment on the assumption that you meant there isn't demand on behalf of businesses for 10X more content. But you may also be saying there isn't enough demand to consume that much content.
In that case, I would argue there's always an appetite for great content. Our expectations just grow as the quantity of available content grows. I don't know about you, but I'm constantly on the hunt for something great to read or watch, and most of the time, all I find is mediocre. I might read one thing a day that really makes me think. If 10X more content means 10X more think-worthy content, stuff that engages me, then I certainly have an appetite for that.
What I do not have appetite for is 100 articles saying the same thing in slightly different ways. If that's the kind of content you're going to produce with the help of AI, then I agree, there's no market for that.
Interesting. I'm an entrepreneur too and the key thing here is people. Less people working = less purchasing power = less money available = less money in our pockets. See how that works? Everyone is going to be impacted and no amount of automation from our part will solve that.
115 comments
[ 2.5 ms ] story [ 189 ms ] threadTaking control of $100,000 robots to manufacture does have value at the high-end, but still can't compete with humans at the low end.
I wonder if this can bring some human-friendly Manufacturing back to certain locations.
The environmental impact of outsourcing is incredibly large, making things closer to home with newly freed hands is probably better.
It might also help with the obesity crisis having people spend time sitting 10 hours a day less frequently.
In other words, the new work fill the same time, and we’ll just be doing 10x of it (if we can solve the current challenges with LLMs). The human mind is unsurprisingly adaptable to changing conditions and great at abstract thinking, which is what ironically distinguishes it from these machines.
We also tend to severely underestimate our own abilities and overestimate the importance of the tools.
Even on HN, this is rampant. I wonder, is it because people are so used to interacting with most other humans solely through a chat interface that they perceive ChatGPT as being close to a real AGI, almost indistinguishable from human intelligence?
Ironically, we tend to limit our imagination on what _we_ could do that computers won't, while simultaneously overuse it to imagine what they _could_ do but won't.
One very good example of this is humor - LLM models at present are terrible at generating anything but dad jokes, and are even worse at their attempts to explain "deep" humor. Gary Marcus had a recent episode on it, very interesting insights from top experts in the field: https://podcasts.apple.com/us/podcast/can-ai-make-you-laugh/.... They make a case that LLMs were specifically optimized for text generation and not explanation, while humans are almost the opposite.
I believe somewhere in there is also our tendency for confirmation bias and overlooking imperfections where we don't expect some. AI may generate nonsense details, but when we look at the overall text, we don't notice them since we're just awed by the output, and think if it looks flawless it must be flawless. Same thing as with mentalist tricks.
And finally, we only think of our intelligence as thoughts carefully animated into sentences, while there's this whole "iceberg" level of feelings and unconscious cognition, which we don't even notice ourselves, but it would be extremely hard to reproduce in AI and we have not even the slightest idea how.
The problem with robots isn't the physical cost vs a human, it's the task flexibility & cost to program.
This is something that could also theoretically be solved with AI in the (fairly close) future.
Latest generation robotic arms with vision are taking on a greater variety of tasks (e.g. bin/case picking from a mixed tote), more reliably, with a better ability to adaptively recover failed tasks.
The progress here are about not needing parts/items to be in a predictable/consistent location and still understanding how to handle/retrieve them and perform actions on them. This stuff is in sites right now.
For your kids, I think you’re imagining more progress with AI than we’ll have in the next five years. I think the changes of the last 3 years will be much more impactful on new grads. I recently switched role and learning remotely is a lot more difficult for various reasons. I’m already well into my career so I imagine for new grads it will be incredibly difficult. I’d worry more about that than AI provided they’re going to college to study something “useful”.
99% of writing is meh, so ChatGPT is not a downgrade.
In addition, I think we’re living in a huge bubble and are sort of entitled enough to think that Internet = human knowledge. So many jobs and skills are simply not transferable even using human language (and even if they hire an army if “trainers”). This is one of the core problems with the current incarnation of “AI” - tacit knowledge has always been the stumbling block, even back in the expert systems days when everyone was thinking what we’re thinking will happen today.
So until the machines can somehow gain consciousness, autonomous and nimble movement, the will to live and all of the deep levels of the human mind that are outside language, I think we’ll be pretty safe :-) And for now, on all of these parameters they aren’t even close to a cat.
Between social sites, corporate BS and medias farms, it's more like 99.9999999999999999%.
Why? Traditional family unit is over and generations are at war. Let them figure it out. Meanwhile let's not have kids without a clear plan for their future.
* Past performance is not indicative of future results.
It's not surprising that the human race has survived for this long - we always find ways to adapt, no matter what challenges history throws at us. This is what distinguishes us (and I suspect will always do) from the machines.
What now? Established trends have been regularly violated over thousands of years.
> It's not surprising that the human race has survived for this long - we always find ways to adapt, no matter what challenges history throws at us. This is what distinguishes us (and I suspect will always do) from the machines.
You know what would count as an adaptation? Large fractions of population becoming immiserated and either dying off or living in poverty at the margins.
Mere survival is the wrong metric.
However, if you'd like to be pessimistic on it I won't argue, I have that tendency too ;)
And to lighten up the mood, I highly recommend listening to this episode with Gary Marcus, where he invites experts in the field of humor to discuss how distinctly bad the LLMs are at explaining what comes naturally to us: https://podcasts.apple.com/us/podcast/can-ai-make-you-laugh/...
"This AI" seems perfectly capable of resulting of extending the kind of economic immiseration that has already affected many blue collar jobs into many types of white collar work (ironically enough, the kind of work those blue collar workers were told to pick up).
Past this iteration of AI, who knows if the advances will peter out, but it's foolish to assume things will be fine because other things were adapted to in the past. That's kinda like assuming you won't die before having kids because none of your ancestors did.
List them.
You mainly been making vague, unsupported assertions backed by emoticons. It's a stretch to call that an "argument," and arguable you haven't been really engaging from the start.
I don't think you can convince me because I suspect you don't really have any support or specifics for your position to provide.
Regardless, I flagged your comment since it really doesn’t follow HN etiquette on respectful arguments.
Come on. If you have them formulated, you should have responded with them. It's not that hard.
> Regardless, I flagged your comment since it really doesn’t follow HN etiquette on respectful arguments.
Again, come on. The disrespectful thing is expecting someone to go on a scavenger hunt just to find out what you, some internet rando, should be saying in your half of a conversation thread.
It's also just bizarre for you to note you flagged my comment. Am I supposed feel intimidated or ashamed or something?
I don’t think that’s true. The archetypal constructs of human cognition have been unchanged. Hence the enduring relevance of myths. It’s always the same stories playing out.
> I don’t think that’s true. The archetypal constructs of human cognition have been unchanged. Hence the enduring relevance of myths. It’s always the same stories playing out.
I think you misunderstand. I didn't say everything has changed, just that many "established trends" have been.
I mean, at some point it was a well established trend that all humans hunted and gathered for their substance, then that trend was violated.
At some point, the trend of "we will find something productive to do with all these people who's jobs have been made obsolete" may be violated. And it seems like that's getting more likely the more rarefied the remaining valuable skills become.
> Mere survival is the wrong metric.
People don't have to live in cities. Knowledge work has brought many people to cities and the absence of it will reverse that trend. It is welfare that keeps impoverished people living in cities.
But I'm basing my opinion on history as well. The focus will be on reducing costs, which means eliminating as many workers as possible.
They don't, and in 10-20 years corporate leaders will bemoan the shortage of high-skilled experienced people that they created through their own decisions. Of course, they won't take any responsibility or change their approach, though.
It's a free rider/prisoner's dilemma problem.
If everyone cooperates then we all come out ahead. If you all cooperate while I defect then I end up even better (I don't have to train anyone). But if we all defect then we're fucked.
It used to be that there was entry level work that needed to be done by humans and so you didn't have this problem. If the entry level work can be done by AI then you need something else. Either government coordination and incentives to hire humans (this is the kind of things gov't is good at). Or people stay in education for longer/education becomes different.
This. Anyone who thinks that their job is safe is fooling themselves.
Is it blindly optimistic to assume that people will eventually yearn for more genuine interaction, daresay truth?
Is it wrong to think that people gain purpose from other people, and that our mammalian instincts will not be so easily conquered?
On top of that, the biggest unaddressed problem that everyone seems to pounce over is tacit knowledge - it cannot simply be transferred even between humans, using words, gestures or any other media. It can only be gained through the human experience, by doing stuff. It’s the same old problem from the expert systems days. So we cannot hope to replace them with AI, at least not until AI somehow becomes self-aware and gains all of the non-surface depths of the human mind, and lives through the same experience.
If your argument is that there won't be anyone to spend money, so even the jobs that AI can't replace won't be safe, then okay.
Maybe Netflix can be less convenient so Blockbuster won't have such a hard time competing.
Personally Id rather have more people dog sitting and fixing AC units, i.e things that provide actual value.
This is horrifying.
I have one friend who, instead of firing her, her company asked her to use ChatGPT and increased her work load. Smart managers will augment their workers, unless, of course, the worker was poor-performing to begin with.
Automation should let people work less. Not just give rich people leverage to work us harder. Making people juggle more tasks at once with the AI will just burn them out.
A bit over 40K if you count everyone, and closer to 60K if you just count people working full time.
The government will later offer debt relief to those who agree to be permanently sterilized. It will offset your debt by giving huge amounts of carbon credits for the emissions your children would have been guilty of.
What an odd statement. Knowledge work has been automated for as long as I can remember. Compilers, automated QA, data entry, etc. In many fields, more labor demand resulted from productivity gains (software); in others, demand didn't rise to compensate (bookkeeping).
The question is where is additional demand going from higher savings given productivity boosts. I don't see more demand for HVAC technicians. Maybe dog walkers.
But ChatGPT is now killing jobs that are solely about creativity, like copywriting. That is a new, and rather unexpected, turn of events. I saw a good quote, I think it was on Twitter: "What kind of world have we built where the humans slave away at physically demanding jobs for minimum wage, while the computers write poetry and paint all day."
But the point is that the kind of copywriting that CAN be replaced by ChatGPT is just as repetitive and non-creative than those other jobs.
I've worked with good copywriters, and was impressed with the quality of writing they did. The point is that ChatGPT excels at the type of creative writing that previously we thought was solely the domain of humans. I think you are getting things backwards if you think that because ChatGPT can do it that it is repetitive and non-creative.
Actually when you think about it, even clerical work was "automated" by printing press.
Automation of production of generic content is bit scary, but it might free up resources or just show what the truly value of that work was.
I would really be worried right now if I were a marketing copywriter. I've seen (and posted about) examples where ChatGPT still falls behind human ingenuity, but the fact is that it really only falls behind some of the best of human ingenuity. Just guessing, but I'd believe that ChatGPT is as good as or better than 75% of copywriters, and even for that upper tier of copywriters, probably only about 10-15% of the writing they do is in an area where the additional quality makes a difference.
Just a worrisome future ahead in my opinion.
What the applicant writes: "I want this job."
How the application is sent: "Dear Sir or Madam, I am writing to you today to inquire about the position of blah blah blah..."
What the recruiter reads: "This guy wants this job."
I'd like a tool I could point at a modern tech company website to read through all the meaningless ad copy and buzzwords and spit out a two-sentence summary of what the thing actually does.
As a demo, try this prompt; you might have to trim some to stay under the token limit:
Summarize the major arguments and sentiment of the following Hacker News posts. Provide your answer in a brief, bullet-point format:
<paste text of HN comment thread here>
They also don't fail to realize when you ultimately drive 80% unemployment due to AI replacing everyone's job. That war is going to be internal and it's going to look like the French revolution.
80% unemployment managed with machine guns and state control of all information sources.
There comes a point at which the plebs can't do anything about their fate.
But that's why we have TikTok, Instagram, Netflix, cheap Corn Syrup, and Porn. You keep the plebs hedonistically satisfied while you slowly consolidate your power and protect it with an autonomous army. THEN you pull the plug.
I think that was very intentional. Put the focus on scary-sounding things that aren't actually an imminent risk in order to distract from those things that are.
I'd be sad to see her lose her job, but at the same time, I can see how unfulfilling this job must be for her. She often feels that the work she does doesn't matter to anyone.
To me, solution is not to invent more bullshit jobs. We need a better system for humans to share the fruits of millennials of progress together.
I'm seeing government outright banning LLMs from certain industries just to avoid social unrest.
The current model seems to be just like Uber. Get into as many markets as you can. Provide the service below cost. Then jack up the price and make money.
As someone who uses ChatGPT as a learning tool for various programming tasks, I'll miss it if it becomes too expensive. Its a helpful tool, but not worth more than $20/month.
I hope and trust that they will find meaning in a career fixing and building and caring for physical things that people actually want to pay for. This is a step forward.
For content to have any value, it should contain unique information or an especially good and useful presentation of existing information. Otherwise making that content is just worthless busywork. Like the example in this thread of a poster making a press release. How could AI add any information that is of any value?
The same goes for AI generated images. If your images are not extremely relevant you are doing it wrong. Use images of your product, not random people smiling. Stock photos, Corporate Memphis art and AI images are a negative value for the information you're presenting.
Just as you, I hope companies some day understand that their worthless filler content is just getting in the way of their sales.
An absurd, but seemingly likely, end state of this is an arms race between the two technologies, completely transparent to the user, and resulting in roughly the same user experience, which is always set to the limit of their tolerance.
I don't doubt the threat to copy-writing and similar jobs, but I also suspect that the call/response nature of the advertising industry is going to result in people being inoculated to ad-copy written by ChatGPT, and it will start to sound as hokey as old ads you find in computer magazines.
For example, today it helped me write a press release. We had dialogue back and forth. I told it to ask me thoughtful questions, and it did. After 5 minutes of back-and-forth, I had the bones of a high-quality press release put together. I then edited it until it met my standards, and submitted it.
This process would have previously taken me a few hours. Today it took less than 30 minutes.
As a result, my value to my employer has gone up because of ChatGPT. I'm able to get a lot more done with it, and they'd be fools to simply replace me with the bot.
In other words, and I'm hardly the only one saying this: ChatGPT is not a replacement for people. It's an augmentation. It's a bicycle for the mind, to steal Steve Jobs's famous metaphor.
I think if you're a writer who refuses to embrace this new reality, you should absolutely be worried. But if you embrace the inevitable, the combination of a skilled copywriter + ChatGPT is incredible. It's my writing partner. I'm still in charge, calling all the shots. But my writing partner helps to prevent writer's block.
Often, I won't use a single word it produced. Instead, it gives me a starting point. Sometimes that starting point is little more than "I need to do better than what this machine produced, because this is crap." But it's soooo much easier to improve on garbage than it is to create out of the blue.
So, I don't think AI is coming for our jobs as writers. If used well, it's the best writing partner you could ask for.
This is the analogy that jumped out to me as well. LLMs seem to be more of a vehicle than a destination. Like a bicycle, LLMs need to be operated with the direction and effort of an operator to ensure it stays on course and doesn’t veer into traffic.
Continuing the analogy, I suspect that LLMs will open up traditional white collar jobs to more people. More people can get to the destination on time with the aid of a bicycle than on foot.
...or we can be freed to spend our time making the content better than ever, because we didn't squander our whole day on the first draft.
All businesses are in the business of maximizing profits. If they can be more profitable with fewer writers, of course they'll reduce staff. But if X content produces Y results, one should expect that 4X content will produce 4Y results. Profitable activities get invested in. Cost centers get optimized.
ChatGPT lowers the bar for content creation, which actually raises the bar for producing content that will stand out. If you're just copying/pasting content from ChatGPT, you'll have a lot more content, but it won't stand out. It's content for the sake of content, as opposed to content that's meant to engage and persuade a person to spend money on your products.
This is especially true with LLMs. They can't come up with truly creative content. LLMs, by design, return the most average content imaginable. This has tremendous value in the hands of a skilled writer/editor.
Remember: The purpose of marketing is to make your brand and your products stand out from the crowd. You're not going to achieve that by producing what is, by definition, the most average content possible.
ChatGPT makes me a better writer. That's all there is to it.
Businesses that don't believe content can sell shouldn't have any writers at all. Those that do should invest in AI-competent writers more than ever.
I maintain that AI is only a threat to me as a writer if either (a) I'm a hack who produces low-value content, or (b) my employer is stupid. I'd rather not work for stupid, anyway.
Let's say it costs $100 for a piece of content, like a blog post. (I'm not suggesting that's the right price, I just like easy math.)
And let's say your return is $200 in sales, leaving you a profit of $100.
If I could double my money like that with every article, I would gladly pay $10,000 for 100 articles if the return is then $20,000. That's a money-printing machine.
Now let's say the return isn't $200 per article, but $105, for a profit of just $5. I wouldn't pay a single penny for it, because I have better ways to invest my money.
This applies to anything in business. Whether it's content, technology, inventory, people, real estate... whatever. If the return is great enough, smart business owners will want more of it.
It has nothing to do with how much time it took or how many people were employed.
It has everything to do with how much it cost versus what the expected return would be.
That's all business is. Turning $1 into $2.
Show me a process that does that, and I'll invest every dime I've got.
Now, let's go back to the original example: $100/article for $200 return. If ChatGPT allows my writer to produce 10X as much content at the same cost (which I doubt... it's a bicycle, not a jet plane, but I digress) without any loss of quality, I would anticipate my returns to be $2,000 for every $100 invested. That would be incredible. I wouldn't just invest every dime I have but I'd max out every dollar I could borrow as well.
BUT: That assumes each piece of content can be just as (or more) effective as those that came before. There's a limit to how much anyone's audience really wants to hear from them.
The answer to this is to invest in finding new audiences. New niches that would want your products. High-quality expert content is a fabulous way of attracting communities of like-minded people. So when you get to the saturation point with one community, there's an opportunity to find another, and then another, and then another, and so on and so forth.
It all hinges on whether the content is actually doing the work it's supposed to be doing. Things that make money get invested in. Things that cost money get optimized or eliminated.
If I was a writer and I lost my job to ChatGPT, I would take a long hard look in the mirror and ask myself why my employer didn't see my value anymore. Was I just producing content for the sake of content? Was my content profitable for them? Was the quality of my writing so mediocre that a machine did it just as well, if not better? If my work generated profits, why would my employer cut me out? Why wouldn't they want even bigger profits? Aren't they in the business of making money?
--
Now, I responded to your comment on the assumption that you meant there isn't demand on behalf of businesses for 10X more content. But you may also be saying there isn't enough demand to consume that much content.
In that case, I would argue there's always an appetite for great content. Our expectations just grow as the quantity of available content grows. I don't know about you, but I'm constantly on the hunt for something great to read or watch, and most of the time, all I find is mediocre. I might read one thing a day that really makes me think. If 10X more content means 10X more think-worthy content, stuff that engages me, then I certainly have an appetite for that.
What I do not have appetite for is 100 articles saying the same thing in slightly different ways. If that's the kind of content you're going to produce with the help of AI, then I agree, there's no market for that.
I think finding right people and managing them is pain in the ass