The Death of a Technical Skill (2020) [pdf] | Hacker News Reader

The Death of a Technical Skill (2020) [pdf]

john-joseph-horton.com

47 pointsmiohtama46 comments

COVID taught me to take these changes seriously, the hard way.

I told my wife that lockdowns/masks would be at least a two year phenomenon, given how long the Spanish flu took - and our mitigation would make things take longer, not shorter. This had many obvious implications in my life, where I lived for the next 5 years, what jobs/skills I should be wary of vs. focus on, what stocks I should bet on.. etc.

But I didn't behave accordingly with this obvious conclusion I had made. I struggled to actually take the leap and make changes in my life. I made very few stock bets, for example. Also we didn't take the opportunity to move/change plan/etc around real estate.

With AI, I feel like this is a similar moment. My employer has an internal copilot thingie, and it's amazing when trying to use some internal library/service I've never used before. It even integrates with Emacs.

What is obviously true about AI, and how does it affect your life? Curious what other software engineers are doing to prepare for these types of obvious changes.

Deer in headlights mostly. But also diving in and trying out all the things!

> Curious what other software engineers are doing to prepare for these types of obvious changes.

Contemplating a professional change. Maybe I'm being overly pessimistic; people aren't being replaced yet. But maybe it's time for something else for me.

There are dire staffing shortages in nursing homes where I live – e.g., people develop bed sores because no one is available to turn them. I certainly could dedicate the rest of my life to that and it would make a real, tangible difference to people's lives.

It's a scary change though. I am a bit squeamish and I wonder how I'd fare. And it's a big loss in pay.

That's where I am in my AI-induced existential crisis, anyway.

Nursing homes abuse nurses, nursing assistants by hiring less number of people, also pay 50% of what hospitals pay. Just beware.

Sounds like an industry ripe for disruption.

Private equity has been disrupting it for a while, that's why it's like this.

Very few industries have been broadly "disrupted" by replacements that hire more people to give more individual attention to each customer. It's rather antithetical to the whole model.

Thanks for the warning.

I'm in small town France. I know the picture isn't rosy. E.g., one of the nursing homes near me is Catholic-run and so understaffed that on Sundays, only those attending mass are helped out of bed.

That place pays its nursing assistants a premium, though wages are still low in the field, compared to everything else. There just aren't many people willing to do a tough, rather gross job.

Toyota is working on nursing robots that can handle many of the routine tasks like helping people out of bed and walk to the bathroom.

That would be lovely, considering the staffing shortages, assuming it doesn't deprive people of their remaining autonomy – when you start doing things for old folks, they tend to lose the capacity to do it themselves, surprisingly quickly.

There's already lots of tech in nursing homes to make transferring people easier on the caregivers, but full automation would be great.

But beyond that, getting people out of bed is one thing. Undressed, washed, skin treatments applied, dressed is another.

How many relatives did you have in a nursing home? People being forced to live there because they typically can't live at home any more, are deprived of social contact. Staff has only minutes to do the most necessary things and rush on to the next. Full automation would be hell. People are social, even if they often don't admit it. Have you ever seen the eyes of an old person getting a gentle back rub and a little conversation? Don't try to take that away.

> Don't try to take that away.

I'm not sure if you meant to respond to me or someone else. Or maybe my response wasn't clear. No problem either way.

I don't think full automation of carer duties is going to come anytime soon. It's far too technical and delicate. And absolutely provides psychological benefits for residents, as you said.

What would be great is if technology could provide even more help for transferring folks from one place to another securely. Lots of elderly people fall. And lots of carers get injured when they try to stop elderly people from falling.

Edit: Maybe it's the word "transfer" that's wrong and causing the misunderstanding. It probably sounds like the old people get put on a conveyor belt or something. "Transfert" is the word used in French nursing homes for helping a person get – say – from their bed to the toilet or to their recliner. And it's risky for everyone involved.

Actually I wanted to reply to you. Now with your explanation I understand better and we have no disagreement in fact. Thank you!

I wish more care givers would learn kinaesthetics (speaking from my own experience). It makes transfers easier for both and you can help people in ways you never thought were possible.

> I wish more care givers would learn kinaesthetics

If you know some good resources, please share!

We have the luck to have some physios specialized to it in the vicinity. In Germany and some other European countries there is the Kinaesthetics Association [0], we contacted them for trainings nearby [1].

Amazing. This send to be a German/Suisse thing? Did not realize that until today.

The principle is to use the bones to carry weight and lots of pivots around natural joints so you don't need so much muscle force (while pivoting only to guide, not to lift).

An eye opening exercise was to sit as relaxed as possible on a chair and concentrate on what muscles you use when starting to stand up. You don't push up for a long time actually, but lean forward until the center of gravity of your whole body is above the rectangle your feet form. Only then you can start pushing up. Try it as slowly as possible and feel every single muscle participating.

So if you help a sitting person, don't try to pull up, but help leaning forward and find the point where the butt starts feeling "lighter". Then the person can concentrate on helping pushing up with the legs.

Also we worked in pairs to learn that you can synchronize your movements by feeling each other and the cared for person can feel the right moment to spent his/her precious muscle force and not work against you. Your tension starts a tiny moment before the actual movement and if that can be perceived, you can literally join your forces.

[0] https://www.kinaesthetics.net [1] https://www.kinaesthetics.de/trainerinnenliste.cfm

I can't imagine another way to make nursing homes even worse off for those stuck in there. People in nursing homes are depressed, hurting, and lonely. So yea, let's replace their care with robots.

Won’t somebody think of the returns ? \s

Robots need not be uncaring

I get what you're saying.

But robots that could give residents safe mobility with less carer intervention would be a big benefit. Transfers – like getting a resident from bed to the toilet – are risky to the resident and the worker; lots of workers get injured trying to stop residents from falling.

Here in France anyway, they're instructed not to try to stop a fall (only to take preventative measures before a transfer), but in reality, almost no one can watch an elderly person in their care fall without intervening.

As I see it, this is another step in a field whose history is a story of force-multipliers. 60 years ago, they used punch-cards. Every decade or so sees a sizable jump in how much an individual programmer can accomplish in a given time frame, and it has only increased the demand for such specialists. For most of that time, we made middle-class money like most other engineering disciplines. We're due for a reversion to the mean. Lately I keep thinking about Star Trek - everyone can issue commands to the computer, but they still had vast armies of engineers.

>Lately I keep thinking about Star Trek - everyone can issue commands to the computer, but they still had vast armies of engineers.

There's a certain truth to that. I have a few colleagues that can't force-multiply with StackOverflow and Google yet.

Their command of English is just not on the level to search the American internet. So they restrict themselves to the German internet.

While ChatGPT works just as well in German as in English and this might enable them to harness it better. I think there will again be some people that are and some people that aren't able to use it properly.

If you can verbally express your issue you can ask it, and then you need to make sure the response actually works and isn't hallucinated, and you will need to make sure it works in conjunction with the rest of the code base.

In essence what we have been doing with StackOverflow and Google is being supercharged now. And while you can find almost every answer on StackOverflow, putting it all together, testing it, designing the system as a whole has taken up more and more mindshare.

Much new software will be created. And all of it will be legacy.

I ask chat GPT to write code and it does it very well. But sometimes it writes real crap. I'm not sure it's yet at a point where you can say, "I have a complex system, I need to keep it running and change it likes this" Jobs that require judgment, change management and handling legacy code will boom.

It's better to think of chat GPT like auto aiming. Or going from hammers to nail guns.

Presumably it would need to be trained with good examples of solutions to complex problems. Maybe there just won't be that many for the foreseeable future.

Yeh exactly. AI was trained on real data, and in a few years, there will only be AI versions of real data to run off. ie. alot of that really crappy code will be deployed to github, and the result will be that and more shitty code will be written.

Thats pretty much what has happened over the last 30 years. You need less skills to produce more and more. Google search started it all, i remmeber back then I quipped the Offshored devs wouldnt be able to get a job without an internet connection because previously us engineers prided ourselves on understand and memorising APIs. The Ai programmer will outpace India's cert mills in quick succession but it wont benifit anyone except the Large Corperations who own the Ai. Atleast the Google devs added questions to SO, the AI dev is leecher who only profits, plagerises, never contributes.

Interesting, point. But won't, there be allot of so called AI whisperers who train and own models that are very good.

> Curious what other software engineers are doing to prepare for these types of obvious changes.

Wouldn't Software Engineers just work on a higher level of abstraction if AI can automate the mundane details?

And by doing so, many people will be replaced. It is undeniable.

Sure, developers will be working on a higher level of abstraction, but in the higher level, you will need lesser people.

So, people will be laid off. Make no mistake of that.

And there will be segments where entire products can be spewed out by AI. There will be 100% layoff on those teams.

A manager can now supervise 5-7 products and there will be one "Chief/Principal Engineer" who does the prompting for 2-3 products. Then there will be systems and devops people with much lesser number.

All of IT people won't lose their jobs, for sure. But many many will be laid off and hiring will significantly be dropped.

And owners/shareholders of IT companies will grow richer and make bank when the techies will be unemployed.

The techies may go on to teach Math or Physics to High Schoolers, or find other jobs, but then those people will be replaced.

UBI will be the way to go. But countries like India, China, Indonesia will be screwed.

Mark my words, very bad days are coming.

Guess that means that engineers should embrace the management side then.

Total numbers of managers will change, too, and the remaining ones will have increased workload.

Working at a higher level of abstraction doesn't necessarily mean we need fewer programmers, because it also means that more projects are economically viable. The software I write now with its complexity and degree of UI polish would be laughably uneconomical with the tools and techniques I was using 20 years ago. Even working at the same company on the same app, 10 years ago I was probably 5x less productive with older tools, yet we've 5xd our headcount in that time despite those efficiency gains. So yea there will certainly be companies that need fewer devs, but there will be new opportunities unlocked(though admittedly the transition process can be brutal on the individual level even when things work out well in the end)

people use to program by punching cards. then compilers came.

I'd say the opposite is about to happen: the boilerplate work will be done by AI, leaving more time to higher level abstractions to humans.

Flash isn’t a good analogy to the current situation with AI tools. But it’s worthwhile to understand regardless.

I think that AI tooling will cause demand for developers to actually go up instead of down. At least in the medium term say the next 3 to 5 years. There’s tons of businesses and nonprofits that could benefit from custom software development, where the cost has so far been prohibitively high. As the cost drops dramatically I expect demand to increase significantly.

What does that cost/demand curve look like? I don’t know. But we are going to find out.

There's also "integration". Outside of the big tech companies are lots of stodgy old companies that will need to connect their legacy stuff to anything new.

While there's an inherent conflict between copyright warrior corporations disrespecting the licenses of FOSS work, AI tooling will be an unprecedented boon for FOSS productivity. It's like giving every lonely cowboy developer a bunch of very skilled interns.

That’s a very interesting point I had not thought about. Let’s say AI tooling results in an across-the-board 2x multiplier for FOSS. Like there is twice as much FOSS. Plus the FOSS that exists today has twice the features plus overall quality. What happens then?

Maybe long-term the main skill in software development is less about writing new code and more about architecting systems. And also using AI to glue components together and debug. Main difference if there’s way more FOSS there’s less need to write new greenfield modules.

I don’t know. I’m spitballing here. It’s fascinating to think about.

it is a problem that whether AI generate a code under copyright. but FOSS developer don't need to worry about this.

GPT-4 can reverse engineer binary assembly code. Everything that runs locally will soon be de facto OSS.

That’s awesome. Have you seen this demonstrated?

I suspect companies will fight this with a kind of runtime encryption.

You said in another thread it's in "the Microsoft paper" -- could you point me at which one you were referring to? I didn't see it in https://cdn.openai.com/papers/gpt-4.pdf , but it's possible that I missed it.

I agree. Large projects become small projects. Bringing custom development to the masses.

I have family in their 60s who will never lose job security due to being experts in Ada.

Apple invited Flash developers at top sites (CNN, ESPN, NYTimes, etc) to a week-long class in embedding Video in cross-browser compatible video codex instead of Flash, right before the release of the iPad, so Jobs would have sites to demo with on stage.

AI is different. A 10x developer can now potentially build at 100x speed with Copilot and GPT4, but with the built-in sandbox it is likely that GPT will not only run but also optimize code beyond what any developer can do. I have no doubt that within weeks if not months, there will AI generated code a human will not be able to understand that will outperform and be available with just a word request.

This is not a shift from Flash to JS. That’s a shift from Software Engineering to being a Business Analyst overnight and the compensation will be even less due to oversupply of white collar talent. Management will be reduced too. This is not doom and gloom, this is prepare for these layoffs to be permanent and increased within months.

> I have no doubt that within weeks if not months, there will AI generated code a human will not be able to understand that will outperform and be available with just a word request.

And are we sure the AI won't be slipping malware into that code nobody understands?

Honestly, you're running it already.

guess skynet in easy mode

Meanwhile, I'm just sitting here, picking bugs out of GPT4 code. But sure, 100x developers.

It's good at well-known tasks. The more you veer into the unknown, the worse it gets. (It's the nature of LLMs). Will it impact CRUD development? Yes, indubitably. But no, we won't have 100x devs, and the SWEs that adopt to using it as a tool will do just fine. We won't run out of work. (One core point is that there's no upper bound for useful knowledge work)

realtors aren't allowed on the moon.

and, who knows...maybe a CDL will still be relevant in 50 years. The COBOL of meatspace.

Reply on news.ycombinator.com