Sam Altman 'A Little Bit Scared' of ChatGPT, Will Eliminate 'Many' Jobs
businessinsider.com
businessinsider.com
It's like using stock photos. I, as a non-photographer, can get access to a ton of content that's of varying quality but all passing some quality bar. The content has none of my context so I need to search/filter/curate. While I, as a non-expert, can do this, at some quality bar that's not going to be good enough and an "art director" or photography expert will be able to do a better job at using stock photos because they understand how to create cohesive selections that fit a brand or brief.
This opens the market dramatically for non-experts, and gives the experts a huge productivity boost by automating leg-work and letting them focus on where they can add value with expertise, taste, context, and so on.
Midjourney might automate those sketches, either for the client or the illustrator, but it's unlikely to automate the final draft as that will contain significant amounts of context that these models are not able to hold, and it's unlikely to automate the touch-up process as this requires the same context plus great care in the adjustments being done, again something these models struggle with.
If this happens, now an illustrator can produce more illustrations in the same amount of time, prices come down, demand goes up. Sure that's a simplified view, but are we saturated on these services? I doubt it in most cases.
How many companies still employ people today to do data entry?
Current non-AI plots don't seem to be especially innovative, so apparently that isn't a major concern of the market.
The idea of a human actors acting out an AI-generated story certainly does have a certain Black Mirror quality to it.
It’ll know how to behave as a cfo, what to look for, and what to reformulate back as responses.
It’ll come for pattern matching jobs first. How long until you can conduct an entire meeting with 5 variants, promoted with different purposes and tasks?
I’m close enough to top management to see how these decisions get made and I have seen 3 instances in the last month of a potential job opening not getting opened because “we could do most of that with chatGPT/mid journey today, and probably do all of it once gpt5 comes out”
The jobs: illustrator, copywriter, operations analyst. All junior roles, all potential career starters that are not going to happen.
Illustrators – midjourney isn't going to replace great illustrators, it's going to allow more companies to have more illustrations, replace some use of stock illustrations, and aid productivity for illustrators in helping them churn out ideas that can be turned into more specific/tailored illustrations.
Copywriters – GPT/etc aren't going to replace copy editors, they're going to allow many more people who wouldn't otherwise hire a copy editor to have better copy, and to aid productivity for copywriters, allowing them to move into more of a supervising role.
Operations analyst – I'm not sure exactly what you mean by this, but if it's business intelligence, there's nothing anywhere near ready yet. These tools don't provide insight, they provide on-demand content that a human can curate for real insight.
In general, these tools will act as productivity aids for the people already in these professions, allowing the jobs to evolve over time. That's not just at the top end but also at the junior end. More people will be able to make use of generated content for free, but they'll all find that it takes time to do so effectively, and find that eventually roles need to be hired for.
If you're currently deciding not to hire these roles because they can be done with GPT/etc already, then I think you're either not quite serious about hiring for these roles, don't understand what goes into them, or have needs that are only marginally above those fulfilled by stock photos/illustrations etc.
But won't it also allow many people who would otherwise hire a copy editor not to hire one?
- Increased accessibility of copy-editing will likely increase the market overall, and therefore increase the market for human copy-editors.
- GPT/etc are good but don't have context, taste, and have proven to be quite fallible over multiple rounds of feedback (common in copywriting/illustration in particular), so offer a good starting point that needs human intervention to use.
- Copy-editors aren't just a transactional prompt-in-copy-out service in human form, they hold products and brands in their head, understand customer segments and how copy applies to them, define strategy, and so on. They are proactive. Companies with these roles, already benefitting from these things, are unlikely to replace them with software (at least not effectively).
The number of roles for humans hits zero.
I think these tools will be usable by people of any skill level, and the result will be commensurate with the user's skill level. You'll still be able to hire junior illustrators, they'll use the tools, and they'll produce better results than if they didn't have the tools, but not as good as a senior person with the tools.
It’s lose/lose.
Instead of having a bunch of humans do a crappy job of writing documentation for your product, which then needs proofreading and editing etc. which requires even more humans, you just keep the one good guy and instead of editing and proofreading and adjusting the output of the humans, they do that w/ the output of ChatGPT.
From what I've seen from our documentation guys, ChatGPT directly employed by our Product people would be a huge boost to quality. As in, we never even sprang for "the good guy" to begin with.
Copywriter: e-commerce company that is constantly creating new SKUs. Managing a human copy-writing pipeline is annoying... many messages back and forth on slack/email between editor and copywriters to make adjustments. Much easier to chat with a computer and ask for more samples.
Ops analyst: different SAAS company main job description was to collate customer requests into a big spreadsheet and categorize, summarize, present to stakeholders. Difficulty comes with scale, breadth of product, communications overhead. Previous hiring plan was to bring on 1 analyst per product line, 1:1 PM:analyst ratio. New plan is PMs do it themselves with help of LLMs.
You do realise you’re going to be expected to check the details in these reports, and your colleagues will all start sending hundreds of “very important” presentations that you must read?
Maybe you can train a swath of other models to synthesize property tests, but then you need to bootstrap this onto itself.
There's a big difference between "No machine could do my job" and "ChatGPT can't do my job."
Most people are saying the latter. Most people are right about that.
I fully believe that there will be some time in the future when we can create programs powerful enough to convert clear enough natural language of business requirements into code. There may even come a time when we can create fully sapient, sentient machine intelligences, smart enough to work alongside us, though that I'm a bit more skeptical about.
That doesn't mean that what we have today can do this. I can believe that there are certain things that ChatGPT and its brethren can do well enough to streamline away a few jobs, but for many reasons, I don't think that will be widespread.
What all of this does mean is that we need to be pushing for political and economic changes to ensure that, when and if the full (or at least much greater) automation does happen, the human beings whose jobs disappear will be taken care of. We need to stop the fruits of automation going entirely to the already-wealthy, and instead make it so people are no longer required to dedicate their lives to toil just to be allowed to survive on this planet.
Of course, we already knew that. See: bullshit jobs, artificial scarcity, etc. But we'll keep on regardless, no matter how many poors have to suffer.
this you or ChatGPT?
First they were 'scared' that GPT2 would destroy the world
Now they will take you out of jobs. The rest of you can become GPT engineers, but we won't tell you how to make models
NotopenAI is as much in the business of FUD as they are into AI
That is assuming this is purely about economics. It may well not be; for example, if art is humanity’s version of peacock tails, then the effort is the point, and if so, then even an uploaded brain of some famous and well-regarded artist will be dismissed as “not capable of real art”[0].
Hobbies for social experience can certainly still be a thing, as evidenced by the fact that art clubs and music clubs survived the invention of recorded music and photography respectively.
Even code has forms of this: challenge devs to write games in 1k, or on old hardware, or code golf.
[0] https://kitsunesoftware.wordpress.com/2022/10/09/an-end-to-c...
Bunch of artist or creative people don't do it "because of the job", they do it for fun and reason that they like drawing, writing, programming etc. There is a lot of artist and writers who live in poverty all their lives. They could stick to their manual labor that provides food and get drunk, watch TV series in their free time but many people still go above and beyond that anyway.
There are of course many people who already have "why go to school, why even bother" attitude and somehow they are still alive waking up in the morning and going to work or whatever else they are busy with.
I don't really attach any significance to employment in the sense of doing work directed by someone else to make profits for some business in order to get more money. It might be a personality trait, but I don't really want or need more money and I hate being told what to do.
However- I have a ton of ideas and side projects and musical and art projects I'm always working on, and those things aren't done for pay, but they are work in the sense that they required sustained, focused effort. And that's hard even if you love what you're doing. And it sure helps if what you are creating is something that other people will appreciate when it's done, and I fear that if an AI can generate something ten times better with no effort at all, then it might start feeling pointless to do all those things. Self motivation will sustain you for a while, but external validation helps a lot after a certain point.
Seems to me like we should reward people for achieving sustainability rather than only rewarding growth.
Maybe it's time to start substituting a bit of competition with a bit more collaboration and looking out for the needs of everyone? Just a thought.
Unionize now, friends. The owners and the workers don’t have aligned values.
I don’t think this will happen on any sort of scale yet. It seems much more likely that big tech uses it to make social media more sticky, autogenerating content (of all sorts) tailored just for you.
I'm already seeing my productivity be 5x.
For genAI, we're barely into its development and there are already tons of applications.
If you're upset about something, obviously all your needs aren't being met yet.
That Culture was a complete utopia from the author's perspective — all their serious drama had to come from their interactions with various outsiders.
Yet there are people here in the real world who find it horrifying.
Those fictional AI would out-class humans by about the degree to which a human would outclass a blade of grass; even if you tone that down a bit and we're just pets rather than plants in comparison, I can see how people may feel that this detracts from their sense of self-worth.
Being fed and sheltered for free isn't sufficient to provide for all of a person's needs.
- Futurama: https://youtu.be/LCPhbN1l024
If it really just expands faster than we can come up with needs, then what is there to worry about?
which means we need to learn terraforming and/or interstellar travel within the next 100,000 years, without killing our entire species while we do it.
Needs are food, housing, and safety. Anything beyond that is not a need, it is a want. We're not giving everyone everything they need, and an increasingly tiny number of people are grabbing anything and everything they want.
We're about to start discovering that in a dynamical system with lots of other intelligent agents, our need for "safety" at a certain point does depend on others need for food and housing.
Let's try to remember that as we are laying them all off to give their jobs to AI.
The line between a need and a want is fuzzy to say the least. To cavepeople, all our needs are met at this point. But we're anywhere near fulfilled and have come up with more terms that enter the 'need' category. Maslow's hierarchy, pop authors like Tony Robbins (certainty, uncertainty, significance, connection, growth, contribution) and Dan Pink (autonomy, mastery, purpose), trained psycologists/psychiatrists like Gabor Maté and Ted Klontz talk about how our relations during childhood reveal a need for bond/connection to oneself and others.
We're not dealing with exact sciences here, so it's not easy to converge on an objective list of 'needs' that are separated from 'wants'. There are studies like the famous Harvard Hapiness study that may give us more objective notions, but you'd need to run several of these across time to find stuff that has stronger predictive value across generations.
We could go on here - I'd argue universal healthcare is a 'need' in order to live a happy/fulfilled/safe human life in the modern world. Many would disagree with me, but the point is the goal post shifts as we progress.
I'd call it a need if the lack of it is correlated with disfunctional states or objectively lesser outcomes (which outcomes are necessary?). Even if you have this clear cut the question of what's considered necessary is an opinion. The implication of this in the conversation is that AI isn't likely to fulfill our needs: we'll come up with more and shift the goal post just like all our ancestors have done until now.
Never mind the fact that for them, the extra does nothing to change their lifestyle, and is essentially just a dollar-denominated high score, while for the rest of us, the smallest fraction of that wealth would be life-altering.
This is not a statement about those individual people’s skill levels, but simply a reality of the constraints applied to them by their organization (largely following prewritten scripts)
Sorry, but this is the last person to listen to about these things.
With just a few minutes of experimentation, it becomes really trivial to see how ChatGPT can accelerate some types of jobs sufficiently to justify reduction of staff.
It will likely create other jobs too, which is also mentioned in the article.
I would have agreed with you two years ago. But knowing what we’ve done with AI before this (grand master level chess, cars that can self drive, etc) building an ensemble from those and what these LLM are able to do, I think if you don’t see 10 years from now being on its head from today, I think you might have lost the capacity to dream.
I don’t think LLM alone are sufficient. But pointing to what they produced in beta and saying “time is static the universe is done AI is a crock” seems myopic.
In fact, if we're so convinced this particular LLM is so great, we should just be chopping it up into fit for purpose agents.
It’s almost never going to replace our jobs, because when we test it against ours we can quickly see gaps and flaws, or worse, its overconfident yet incorrect answers.
But when we apply a surface level understanding of someone else’s job to a surface level investigation of AI’s ability to do that job, we keep coming back to the same answer: it’s here! It’s ready!
Meanwhile, if you go ask those other people, they’ll show you a ton of very good reasons why AI isn’t ready to replace their jobs, things I never knew they had to do or understand. And I bet a few of them will tell me that they heard it can write code.
There is a possibility that what he said here will turn out to be true. But compared to writing or electric power etc? Surely we will have to wait some time before we can say that. I don't think he is a fundamentally dishonest person so it's unsettling to see him say this with such confidence because it shows that he has lost some grounding in reality.
Wolfram has a tendency to follow the principle "I would have written less, had I had more time" but he didn't, so he didn't.
"Honestly, I'm too hard-working."
chatgpt only knows about the 'open source' world, it only understands one generic human as an audience, and it is absolutely atrocious with details.
it doesn't know how to deal with a bunch of obscure proprietary niche-domain one-off systems, lots of different human personalities and their communication styles, and it's attention to detail is pretty atrocious.
so it would be absolutely terrible as an employee in most workplaces.