Ask HN: How to Prepare for the AI Apocalypse?

20 points by mouzogu ↗ HN
I feel that AI will do to ${insert-job} what Uber did to taxi drivers.

There will be much resistance at first, but eventually the convenience and profits will win out.

And recent history has shown that increases in productivity & efficiency doesn't translate to increase in prosperity (for most).

How can I position myself to not get left behind in this coming paradigm shift, and it is one...i've played with chatGPT long enough to know this sh1t is real.

56 comments

[ 4.8 ms ] story [ 135 ms ] thread
Get rich before it hits.
I think that excellent advice, especially for someone going into a high paid field that could save most of what they earn.
(comment deleted)
As a plumber or electrician you should be irreplaceable. Probably as electrical engineer with a need for occasional soldering in a lab too. Also as a founder of chatGPT based outsourcing company that does the job of 100 with 10 employees you should be safe.

Funny enough my classmate started chatbot company few years ago. I guess he’s doomed with all his investors and will be off market soon enough.

I know electricians who easily make as much as software engineers and yet there is a still a lot of stigma around them as lower paid blue collar workers.
Being salaried electrician things are much worse than being self employed. But being self employed monetarily is close to software development freelancer. In Germany it’s easy with client acquisition - after receiving license you are being added to official electrician registry for your region. People just call through registry and ask who can come earlier than 6 months. Demand is huge. Solar panels can be installed after 18 months.
I’m interested to know other peoples thoughts about the validity of the threat. However, I think that if it is true, a focus on being a competent and positive employee will be a fairly safe hedge. The future of companies will not be your boss and only robots. Plenty of technically minded AI interfacers will be required. That could be you!
It seems like a valid possibility, but no threat in the here and now. I’m not dismissing the potential. I have played with these models and I am gobsmacked. But fundamental capabilities like being truthful have not been unlocked and we are many years of dev and many billions of investment away from an AI model replacing the first knowledge worker in a net loss for labor and a net gain for capital.

Like with self-driving cars, they are irrelevant research projects and convenient safety features that don’t really change anything, right up until the moment they put the very first level 5 car on the road, then the old world passes away and we will live in a new world.

Well, they have already put L5 self-driving cars on the roads and the result is pretty dismal (at least in SF)
These are not level 5 as traditionally defined. They are level 3 vehicles with a broken safety driver that sits in a call center and can’t actually control the vehicle when they get into trouble.
Still going to need people to craft those instructions to the AI.

What do we do today? Craft instructions for a computer. Not much difference if a computer is on the other side, or if an AI is.

Going from assembly -> java didn't eliminate all the programming jobs, AI won't either.

(comment deleted)
> What do we do today? Craft instructions for a computer. Not much difference if a computer is on the other side, or if an AI is.

The difference is that the skills required to craft those instructions (prompts) will be diluted, so anyone can do it. The software industry will become like the news industry. Very few will survive what's coming.

There’s a balance between capital and labor. During some periods labor is strong and you get unions, 40 hour workweeks and OSHA. During some periods capital is strong and you get child labor and stoking the blast furnace in your underwear.

AI is part of the power of capital. If it effectively replaces large amounts of labor, you will see that balance shift to capital having most of the power in the world and labor having very little. There would still be jobs for most, just possibly very badly paid ones. There will always be a price where human labor is cheaper than AI labor, even if it’s a few cents an hour. If the price of labor goes down enough we will be like serfs again, doing backbreaking labor for the day’s bread.

So the simplest recommendation is to have capital, that way if the value of your labor crashes, you would still have a productive resource.

Or ya know, vote if you still have a democracy, and pick up the pitchfork if you don’t.

(comment deleted)
(comment deleted)
I have been thinking about this a lot lately, and I don't think that AI will replace programmers soon.

Software is an unlimited space, sky is the limit to what you can accomplish. AI will be available to each and every company, so every company will try to compete by adding more features exponentially as AI makes adding them easier.

As a programmer your role will gradually turn into more of an architect. You will put things/systems together and make sure everything works, and you will focus less on the coding part as AI does that for you.

but how many architects are required? and it's also (possibly) assuming that this AI will not get a lot more advanced - i believe what we are seeing is just the tip of the iceberg.
On the other hand, programming has only gotten easier over the years and yet there are more programmers than ever. The ceiling of what's expected gets raised as our capabilities rise.
(comment deleted)
a lot of the skills that one needs for a job that we're arguing is beyond the AIs, like software architect or writing anything with real thought in it, implies that we humans are smarter and we can just do the more interesting writing. How does one become a good writer? By writing a lot of stuff at a lower level. I am concerned that by putting all of the grunt work on the AIs will remove the training opportunities that the humans need to actually do the higher level stuff. Removing a few rungs at the bottom of the ladder of becoming better at your career.

I became concerned at this after reading an article a while back that talked about how some sports writing has been automated for years (how hard is it to write that this team won, the other lost and the score was blah) but that work was needed for writers to get to the next level.

A similar thing could be said about technical support folks. The AI can handle the easy ones, but without the easy ones, how do we bring up a new tech to level 2 if level 1 doesnt exist any more? Also, there are a lot more level 1 techs than level 2.

The LLM you played with has no real problem solving skills because it lacks a sufficient model of reality and planning capabilities.

Highly recommend this 1h interview with Yann Lecun to put things into perspective. https://podcasts.apple.com/us/podcast/big-technology-podcast...

I myself learn construction and the trades on the side. In the US this is quite a safe bet. And I wanted to build a house anyway, so why not do it myself and learn?

I’m also not scared of losing my job to a LLM. I do so much detailed problem solving that I wouldn’t mind assistance that 10x my work by taking away all low and mid hanging fruits.

YMMV

I suspect the LLM will end up being just one piece of an eventual general AI, just like the language center of our brain is just one region. You could design a complex of models that cover language, memory, real-time data, sensing, formal logic and multi-media generative output. Most of this could be built today by finding ways to recursively feed back in prompts into a set of best in class models.
I've been looking to learn some trades in my spare time. Do you take any courses or just youtube? I've been interested in finding some hands on courses, but most I see are online prep for some sort of exam
I use a parallel approach:

1) I bought a vacant piece of land that is legally buildable. This is my starting point to build a house. A 5-10 year project. But, I didn’t learn programming in a day, so these seem to be common timespans for learning complex skills (welding, carpentry, paving, concrete pouring, …)

2) I take courses at maker labs which have heavy equipment and then build all the small things to better get to know the machines. Kitchenware, furniture, cups, etc. The less prefabricated parts the better. My next picture frame will be made from wood I cut on my property.

3) I share with the world honestly, including my stupid mistakes and all the layman stuff I do. The feedback I get from experienced people seeing me struggle and having the need to chip in is pure gold. Saves me a lot of time and money as long as I can take their criticism. I have a thousand teachers (I’m exaggerating but you get the idea).

My channel, if you want to see examples: https://youtube.com/@danrl

(comment deleted)
Learn to build/use AI professionally?

The only way to avoid suffering from a change is to be a part of that change.

For me that's translating into spending my weekends to look for ways to monetize state of the art AI(I know, problems before solutions), and to learn about the theory 1-2 hours a day.

IMO we still have at least 2 years to prepare for a huge economical/societal shock. About how long it'll take for smart people in automatable fields to start to transition to tech

(comment deleted)
Think about what it must have felt like going from statically typed languages to dynamic languages. Or even something like the jump from assembly or fortran to C.

In fact you can go back and read some of these accounts. It felt to them like the end of programming, that any lay person could now do it. One generation passed and then everyone just understands that to be what programming is, and assembly seems like some kind of ancient wizardry done by the ancestors.

This is what is going to happen with AI assisted programming. People will still resist using it, the way people resist taking a few months out to learn how to use python. Most people are also not that interested in taking a few hours out to learn how to program with chatgpt.

For people who are willing will become the new programmers, A stakeholder will ask for something and you'll turn around and ask for nearly the same thing to an LLM or other model. It seems easy but so does googling, AI is like one step easier past that - and for now you still have to make minor corrections like you would with a stackoverflow post.

The moat on programming is not difficulty. It's been quite easy to learn programming for the last 5 years or so. The moat is the desire and interest to learn. As long as programming still seems fun to you, this is just what programming is now.

Instead of worrying, learn to use it to extend your abilities early and you'll have a huge edge. Many people are still fighting AI tools. People fought Linux, hard for about the first 10 years of it's life before it started getting really deeply integrated into every company. I suspect we'll see some version of that with both companies and individuals inside companies.

Being honest, I am not that much concerned with what can I do as individual. If sh*t hits the fan we’re going to have to solve this at societal level, not individual.

As others already put out there: if society collapses heavily due to displacement of labour, money won’t matter much.

Unless advanced AIs are used to build weapons for those who own the capital, nothing can hold off angry mobs everywhere.

When we get to the point where human labour in general can be easily automated, that’s when capitalism like we know must be buried. We must find another way to live and prosper on this planet.

I don't get what use will a potentially cushy job be if society around you is collapsing. So you're either wrong about the AI apocalypise, or if you're right, it doesn't matter what you do.
The thing about Uber, is it's replacing taxi drivers with... other taxi drivers.

chatGPT is still just Eliza amped up 1000 times. it isn't intelligent. it isn't aware.

This too shall pass.

This is a Microsoft blog post, so grain of salt sure, but isn't GPT-3 already being used by some companies?

https://news.microsoft.com/source/features/ai/azure-openai-s...

Does it need to be intelligent to make an impact? Human review still heavily used in this Carmax example, but still, important and useful? Would Carmax really go back?

In my personal life, I've used it for help with writing. The only thing that'd make me stop is a paywall, which admittedly may happen due to the cost each query takes.

Lucrative profession: scrum master for a team of chatGPT instances.
Andrew Yang said we needed a UBI because AI was replacing jobs any day now. Then we had a massive labor shortage.

ChatGPT hasn't changed anything. If it was truly groundbreaking jobs would be knocked off left and right. Instead people play around with it and then move on. I still have a team of graphics designers - wasn't image generation AI supposed to replace them? And when are self driving cars coming? I wouldn't worry.

"The Segway PT made a splash in December 3, 2001 when it was unveiled on Good Morning America. Inventor Dean Kamen said that an urban transportation revolution was coming and cars would be obsolete. Kamen felt it was absurd to use 4,000-pound cars and trucks for short trips."

20 years later: https://www.cnn.com/2020/06/23/tech/segway-pt-shut-down/inde...

One thing that's interesting to me is the AI Apocalypse is dependent on substantially large scale ongoing chip production - in particular GPUs. Chip production supply chains look a little more shaky these days for both logistical and geopolitical reasons. So another possibility I've been considering is that AI development outstrips the world's ability to supply the necessary chips at some point in the future. It's not the highest probability outcome at this point, so I intuitively ballpark the chances of this outcome at 20% in the next 20y, but I haven't done any hard analysis.
In 2018 I attended one of Yann LeCun's talk, one person basically asked the same question than you, more precisely the question was "what advice can you give to young people/high school students who are about to start choosing their career?".

He explained that AI will mainly replace "basic" jobs like forklift operator or any job that needs repetitive tasks. Simply because these jobs have no value. On the other hand, he said that at least two kind of jobs will survive : jobs where you create value (like artists, creator etc.) because people will always desire something that took someone's time and attention to make. And jobs where you take care of others (teachers, nurse etc.) because no one want to be taken care of by a machine (read Asimov's novel The fun they had on that matter).

80% of current office tasks could already be automated with a simple script. Tools will only replace jobs when there are people that can understand and implement them. Companies are not efficient.
I don't think there's going to be apocalypse. The situation is bounded by non-tech activity of human. Human still eat, sick, stress (needs entertainment), socializing (buy stuff to show off). There's going to be tech jobs to serve all these things for sake of convenience enabler. AI is the engine, not the applications.
Some aspects of programming are rote translation of human language to a machine language. That's relatively easy to assist with a statistical predictor. Assume that will happen, and get embedded into your editor.

Most interesting aspects of programming involve deciphering vague ideas, modelling processes, correcting faulty assumptions, presenting alternatives, debugging unforeseen issues, and adjudicating disputes. Those won't go away until we have human-equivalent general AI.

So: move your way up the value chain from reproducing CRUD apps and tweaking CSS to mediating business problems.

> what Uber did to taxi drivers.

And like Airbnb did to hotels? No. "Big business" (in any sector) just needs to outlast.

>> i've played with chatGPT long enough to know this sh1t is real.

It still gets basic facts and primary school math wrong:

https://twitter.com/rasbt/status/1620258064690724864

Language models can't and don't "think". They generate convincing looking text regurgitated from all of the previous text they have ingested.

You can call it what you like — thinking”, “not thinking”, or something else entirely. But LLMs and other generative models clearly have capabilities that we haven’t seen before from AI, despite their limitations.

And there is no reason to believe the tech won’t continue advancing. How quickly, it’s hard to say, but it will happen.

Personally, I’m not too concerned about skilled jobs being displaced, but people should keep their eyes open and be ready to take advantage of new tools as they become available.

>> LLMs and other generative models clearly have capabilities that we haven’t seen before from AI, despite their limitations.

It's an advanced / automated version of the "Cut-Up" technique:

https://en.wikipedia.org/wiki/Cut-up_technique

LLMs and generative models can create some impressive results, but people have been using it and similar techniques for more than 100 years.

>> I’m not too concerned about skilled jobs being displaced, but people should keep their eyes open and be ready to take advantage of new tools as they become available.

Agreed. Calling it "the AI Apocalypse" (from the title) is hyperbole.