AI won't replace human devs anytime soon

https://twitter.com/skeptrune/status/1843060221494895058

by skeptrune • 2 years ago

109 157 2 years ago

I agree but it's important to know that AI is a tool that devs need to understand and incorporate into their daily work life.

skeptrune3• 2 years ago

100%, but my mental exhaustion grows every time I see a post claiming Claude or some other LLM can build your whole company

mattl• 2 years ago

Nobody needs to incorporate AI into their daily work life.

al_borland2• 2 years ago

Is a statement like this any different than someone saying developers needed modern IDEs 20 years ago? Yet we still have people in the industry using vim or emacs, and are successful.

I will occasionally ask Copilot for something when I’m in a hurry, and the there is some minor thing I don’t care about that someone is telling me needs to get done. The other 99% of the time, I’m doing it myself. Any time I reach for AI instead of doing it myself, I’m learning less, growing less, and understanding my code less. Why would anyone want this?

lz400• 2 years ago

I think it's more like the move to cloud. It didn't make developers obsolete but it DID make some types of jobs semi-obsolete like sysadmins and dbas. It also created new jobs like DevOps.

bdangubic7• 2 years ago

once you realize you are 30-40-50..% slower than your peers you will not only want it but need it to keep being employed.

saying “i am learning less, growing less…” is just so amazingly shortsighted, like farmers when first tractors arrived going “fuck this, imma walk this field in 100 degree weather - doing it myself”(sound familiar?).

greenhat762• 2 years ago

I still use vim, has all the features "modern ides" use. It just isn't a gui.

marcosdumay2• 2 years ago

Well, it's actually different.

The impact of the "modern" IDEs of 20 years ago was distinctly measurable, and sensibly perceptible by anyone that tried them.

At the same time, it's not even clear if the gain from LLMs is even positive.

But if your claim was that neither was relevant enough to change the overall noisy productivity of developers, than yeah. At least not positively.

grayhatter2• 2 years ago

funny, I recently switched back to vim, and delayed enabling many IDE features such that I don't even use tab complete anymore. I'm actually enjoying writing code again. It's easier to get into the flow state, and generally I find the code I do write to be a higher quality. While the last bit might be completely subjective, and likely a bit of not just sample, but also confirmation bias. I have no interest in any flavor of code complete, or other advanced IDE features. For me, it's seems to be a net loss.

I understand how generative AI works better than most of the SWEs I work with, but I have absolutely no desire to incorporate it into my workflow. I like that I understand how the code I wrote works, but even if I didn't, I wouldn't trade this rediscovered enjoyment for anything, including the "promised" career velocity.

dboreham2• 2 years ago

vim would be a luxury for the likes of me. Still using ed here...

add-sub-mul-div• 2 years ago

Not for writing code, no. If you're experienced enough it's going to slow you down.

the_gorilla• 2 years ago

It's important for low skill developers. I never understood why everyone was afraid of the shit chatgpt output, until I finally realized it actually surpassed many "software engineers" in skill. Sobering!

gerdesj• 2 years ago

Your comment is rather prescriptive without any working. Why do I need to ... ?

So far, I have managed to work out how to spot a hallucinating ChatGPT derived "organic blog post" but it is bloody annoying.

Back in the day, Linux related queries would generally end up in Gentoo, Arch, Ubuntu, Mint (int al) forums or wikis or perhaps Reddit and co, sometimes it would end up in TLDP by accident. Now we have a plethora of wankery "blogs" that clog up the search returns. Mostly very pretty and mostly following the same old pattern and mostly correctish and wrong at some crucial point.

I have to deal with a lot of pretty complicated and quite niche stuff, for example HA Proxy, Apache Guacamole, on Linux which itself (despite running mostly everything on our lovely planet) is also considered niche.

Windows related queries normally end up with a post on an MS site with a response that suggests that "SFC /SCANNOW" will fix everything from stiffness in the joints to cancer. That has always been the way since around the late noughties, thanks to a rather lax hiring policy becoming the norm in a large part of the world using internet points as a score for hiring. That is understandable and humans being lazy and abrogating responsibility is not a new thing. Now we have multi zillion <currency> things happening that claim intelligence and what looks suspiciously tulip flavoured.

Oh dear!

bigstrat20032• 2 years ago

Only if it provides value, which at this point I wouldn't agree it does.

fzeroracer2• 2 years ago

Honestly, when I see stuff like this I laugh all the way to the bank knowing how many developers and teams are opening themselves up to massive security holes and/or bugs by trying to incorporate LLMs into their toolsets.

People are just willingly leaving massive landmines across their codebase waiting to blow their feet off and don't even have the experience to know when the code generated by the AI is bad or not.

rsynnott4• 2 years ago

need to understand

To some extent.

and incorporate into their daily work life.

No, not necessarily. Frankly it's of no use, or is actively counterproductive, in many fields.

NBJack• 2 years ago

Honestly, this opinion is borderline tinfoil hat territory, but I'd like to point out that between the sudden market expectation that fired hundreds of thousands of developers across multiple companies (in a very close time period!) and constant "LLMs that replace humans are just around the corner" marketing drivel, it almost appears as if The Market (tm) wants to really just drive down the costs of developers more than anything else.

Perhaps it's all a completely legit market turn, and I'm the one in the dark. But I continue to be pessimistic that LLMs are truly going to change anything for the better.

onlyrealcuzzo2• 2 years ago

There's already a type of search that is much better now than 3 years ago.

skeptrune3• 2 years ago

I made an HN search engine which I think is better than the existing - https://hn.trieve.ai and I still agree with the above comment.

Tinfoil hat vibes all around for me too with the LLM's taking Dev jobs.

makeitdouble• 2 years ago

For people not actively dipping their toes in the field, which type of search do you have in mind ?

majormajor3• 2 years ago

2020 bubble hiring was weird enough that I mostly throw out the 2020 hiring and the 2022 layoffs both and just look at things compared to 2018/2019. That ramp-up in the 2010s was called a bubble for years even before Covid but has never popped yet. The expected product bar just keeps getting hire instead.

mlinhares3• 2 years ago

I don't think it was just 2020, when you started to see people being hired out of bootcamps with 0 experience whatsoever to entry level jobs it was visible there was something weird with the market. That has mostly reverted and I haven't seen a bootcamp ad in a while.

qntmfred• 2 years ago

The Market (tm) wants to really just drive down the costs of developers more than anything else.

well yeah. this wasn't going to last forever.

in the US especially. Salaries here were bonkers compared to the rest of the global developer workforce. Gotta remember, remote work wasn't common and barely possible at scale not that long ago. COVID put an end to that, and with it the expectation that management hire developers in any particular local geography. Despite all the RTO drama in some organizations, the majority of the economy was always going to cash in on the advantages of hiring developers in other countries, and squeezing even more productivity out of them with AI tooling.

Downward pressure has been a long time coming. Hope y'all enjoyed the ride.

JTyQZSnP3cQGa8B• 2 years ago

It’s not your tinfoil hat IMHO. I see a few managers and non-technical people (IRL and more on LinkedIn) who “study” prompting with the idea that they can replace people with LLMs and cut down costs.

LLMs can be useful in very specific tasks but these people don’t care about it as they don’t have the technical knowledge to see how and why. All they care about is more money for them.

It won’t be limited to developers though if that happens, and with the darkest scenario, UBI will have to happen as you can’t have half the population unemployed roaming the streets.

Ekaros• 2 years ago

I always question was the firing due to some LLM or changing market conditions namely much higher rates. As higher rates means investments don't need to search for new products and actually companies investing in products with debt get lot more expensive. Thus there is less demand for new products thus less demand from someone to make those.

It could very well be time period where just less software will be produced. And software that is need to make returns at least in medium term already.

faangguyindia• 2 years ago

80% coders, just make CRUD apps.

Those working in specialized fields will hold on for much longer.

But CRUD apps, apps requiring gluing bunch of APIs with backend and database wil go away.

Most apps do not need much scaling so highly specialized scaling masters aren't really needed

Currently using zed editor, I am blown away by its Ai integration.

Though completion via FIM throut custom LLMs is lacking

And there are other problems like lacking git integration, I prefer vscode for that!

kordlessagain• 2 years ago

So an agentic system can handle git. Would you be willing to use that to manage those processes for you? What about Docker?

selcuka• 2 years ago

But CRUD apps, apps requiring gluing bunch of APIs with backend and database wil go away.

They were already going away with the improvements in no-code tools, Zapier, scaffolding frameworks etc. LLMs are just another automated tool in this context.

majormajor3• 2 years ago

CRUD isn't the complexity with CRUD apps. Working with the rest of the business to make sure the processes get built out in ways that make sense and are actually worthwhile is.

So maybe one dev can handle writing five or ten times as many CRUD services as before. The idea that the biz person is gonna get rid of those devs entirely and babysit the coding agents/deploys/etc themselves strikes me as wildly unrealistic, though. Nor is it realistic that SAAS as a market will just dry up because individuals will just have their own coder agents that do everything for them.

Question is if there's going to be an "enough is enough" point feature/change-wise. But I can't think of the last time I worked at a company where the product team and senior management threw up their hands and said "we did it, we finally cleared the entire product development backlog" as opposed to generating ideas at a rate many multiples of the rate of implementation.

The baseline "here's what you can get with a shoestring team" level of output will get higher. But it will also get higher for your competitors. So expectations will just be much higher. So you just gotta move that much faster now.

swagasaurus-rex• 2 years ago

Crud apps can unexpectedly become surprisingly complicated. Take the humble to -do list for example.

So you can create, edit, even delete to-dos, and it’s even hosted online.

* Once a user has a lot of to-do’s, they might want to organize them. You could organize them hierarchically, or with tags, or represented as a graph.

* what if somebody makes an a new to do or an edit while their cell phone is out of service?

* Are multiple users supported? Can users collaborate on a single to-do? Are entire organizations supported?

* what happens if somebody accidentally deletes a to-do? Can they ever get it back?

* What about concurrent edits to the same to-do?

* Can just anybody make to-dos? What if somebody writes a bot to make millions of to-dos on your website?

* If I load up my to-do list when my phone has no service, do I just get a loading spinner?

The perfect to-do app might actually take an expert a year or more to write. Then it still has to generate a profit somehow.

greenhat762• 2 years ago

I work for a company that builds apps for companies, most of our apps start as "a simple crud app" and turn into massively complex projects. I feel as if the business community overestimates the ability of AI code generation. It'll get better and better, but I don't see it getting good enough not to have a developer cleaning up its mess behind the scenes.

015a3• 2 years ago

Those working in specialized fields will hold on for much longer.

The thing about AI up to this point is that it has replaced, well, very few jobs, but what jobs it seems to be most capable of replacing are extremely counter-intuitive. Eight years ago, everyone thought it would be: self-driving, machines, hard labor, etc. Turns out that stuff is really hard, and the first industries to fall were actually the more creative ones like writing and art.

If you have an intuition for how the next ten years will look, I'd implore you to be open to the reality that there is no way you could have predicted the world-state of 2024 from the perspective of 2019; and that's only 5 years.

Here's the counter-intuitive take that I believe to be true: Specialized coding might fall faster than generalist CRUD apps, if either falls at all. The value in specialized engineering is biased a lot heavier toward knowledge of the specialized thing you know about. Versus, the problems generalist CRUD coders deal with every day aren't actually technical or coding problems; they're business problems, coordination, resource allocation, and politics. AI has demonstrated itself as being pretty good at knowing things, even highly specialized things; it has not demonstrated itself as being very good at taking responsibility for its actions.

twelve402• 2 years ago

the first industries to fall were actually the more creative ones like writing and art

have writing and art really fallen? aside from stupid SEO drivel writing, that is

charlieyu1• 2 years ago

We just don’t have a massive breakthrough in robots at a level comparable to LLMs. Which is kinda weird because there would be of even higher demand than language models

krater234• 2 years ago

Both are technical problems with technical solutions that have to fit 100%. You can't say this for art or writing. It's more mushi and anything can fit there.

123yawaworht456• 2 years ago

95% of real dev work is digging through piles of brownfield spaghetti to fix a bug or implement a feature. human-provided descriptions of the bug to fix or the feature to implement are often so vague that even a human developer can barely comprehend them. also more often than not, a visual link is required between the developer and the product being fixed/improved.

hypothetical AI will replace humans, sure. LLMs probably won't. original GPT4 was the best coding LLM I dealt with, and it was 1.5+ years ago. the context size was pitiful, yes, but it was a lot smarter than the current crop of big models.

015a3• 2 years ago

original GPT4 was the best coding LLM I dealt with, and it was 1.5+ years ago. the context size was pitiful, yes, but it was a lot smarter than the current crop of big models.

Anytime someone says "this is the worst it'll ever be" I can't help but think "oh sweet summer child"; and then I ask them, what's your favorite operating system you've ever used?

The answer heavily depends on the person's age. Windows 10 or 7 dominates the conversation, and for slightly older folks, XP. "MacOS Snow Leopard" gets said a lot, that was a beloved MacOS release because they very famously said "no new features we're just fixing the old ones" and they actually did it.

No one ever says "MacOS Sequoia" or "Windows 11". Man, the discourse surrounding the new iOS 18 Photos app is reaching a fever pitch.

Sorry for the weird music on this, but its the first clip I could find of even, gasp, Elon Musk, stating this very real fact [1]: Technology does not automatically improve. In fact, it seems like especially software loves to get worse, and it takes a focused effort, from talented individuals, as far removed from money and business and agilescrum as possible to keep things operationally stable.

But, sure, definitely the new metaversecryptoai bubblescam will escape that and be built to the highest possible quality. There definitely won't be investors and men in suits in every room saying "we're losing billions is there a knob we can turn to reduce electricity usage?" OpenAI is definitely raising billions to fund the R&D of the next level of intelligence, and certainly not because without that capital they'd be bankrupt in three months.

[1] https://www.youtube.com/watch?v=aDC6kWC8y2Y

shepherdjerred• 2 years ago

I would pick Windows 11 (LTSC) over any prior version of Windows, and I would pick macOS Sequoia over any prior version of macOS.

audunw• 2 years ago

My favourite version of Windows was Windows 2000. It was stable, visually consistent and fairly simple.

But Windows 11 is a massive improvement over Windows 2000 in many, many areas. Just because Windows 2000 was my favourite doesn’t mean I think it’s the best. It doesn’t even have built-in support for WiFi.

I think they’re even moving towards something that could be my new favourite Windows. Windows 11 is the first Windows with a visual style that I actually find appealing, and they seem to be making some serious efforts to make it all visually consistent too. The new settings app has its flaws but if they just finish the work of supporting all the features of the old control centre and finally have a unified place to change settings, I would say it’ll be all-in-all better than Win2k.

The GPU driver system and WSL2 are leaps and bounds beyond what Win2k could do. The built in support for virtual desktops and window arrangement is something I would never want to be without.

Yeah, on the Mac side I remember Snow Leopard fondly. That doesn’t mean I think it’s the best. On Mac I see even less reason to think older versions of MacOS is better in any way. They’ve stayed visually pleasing and consistent, haven’t made any huge UI changes similar to Windows’s attempts at making a better start menu, and have added many genuinely useful features over the years.

benterix• 2 years ago

it takes a focused effort, from talented individuals

You see, that's exactly why I don't have high hopes about OpenAI. They came up with a state of the art GPT-4 that was slow and expensive to maintain, so they came up with this o1 that is basically a chain of cheap 4os, something that people used to do from the moment LLMs appeared. And they started calling GPT-4 a "legacy model".

In the meantime Anthropic is steadily raising the bar with each iteration. But even there the efforts plateaued and what we have still has the inherent problems of LLMs, that is the ability to give completely wrong answers with amazing self-confidence.

lokar2• 2 years ago

Even just the ability to reproduce a non-trivial bug, and then work out what general area of the codebase might be in seems out of reach for an LLM

skeptrune3• 2 years ago

Yes! I think I'm going to make LLM bug reproduction my primary response to AI developer maximalists.

sureIy2• 2 years ago

The instructions seem to have to be so specific and repeated that you need a full-time person to review and retry.

My last struggle was creating an SVG arrow pointing to a specific corner. Both Claude and CGPT failed with many attempts.

beeflet4• 2 years ago

This actually gives me a funny idea, IDK if it would actually work.

Something interesting you can do with LLMs is you can constrain their output. What if you just write a test case to encapsulate the bug then try to filter out the most "sane" LLM output that results in the code compiling and the test case passing?

It doesn't seem like much, but it's like using more computing power to filter out all the "obviously wrong" solutions. Doing this in practice might require you to make too many test cases though.

moi23882• 2 years ago

I hate to break it to you, but that’s exactly what ChatGPT o1 excels at.

I swear to god, it’s better than our junior devs fresh out of college.

Like, I let the LLM solve a PBI and it’s just faster with better code.

Actual devs will still have a job, but the low hanging write some code guys? They’re done. And a lot of companies had a bunch of them..

n_ary3• 2 years ago

I hate to break it to you but o1 is like far off of anything what you describe. Yesterday, I was testing some stuff and told it in very precise prompt to write a dripping bucket rate limiting middleware and it just literally named a var “bucket” with a timeout and then implemented something that will wait for the next function to timeout!

I then said, heh and asked it to implement a circuit breaker pattern using axios, and it again implemented a timeout and added it to axios options!

Could be am dumb or o1 was hallucinating on me intentionally, but the thing is very far away from a Jr dev fresh out of college.

yawnxyz• 2 years ago

I had to spend six hours today to untangle an o1/sonnet mess today and ended up rewriting a huge part of it to be remotely readable

what a waste of a nice sunday

floppiplopp• 2 years ago

I've seen a whole gamut of junior devs and their abilities. Your junior devs are just the average beginner. They still learn to get better, to become real devs. And we need real developers to deal with the stuff even o1 stitches together. Because you actually need to understand problems and code to solve problems. And even something like o1 is just a statistical model that outputs the most likely result based on an input. And I've seen the incoherent garbage it babbles out when it comes to code. But the main issue will be, the blind reliance on these hallucinating models that create 'just good enough' will hinder people from learning and becoming better. LLMs are the end of ideas. Just the calcified present. And that's a problem.

raxxorraxor• 2 years ago

I would need to ask ChatGPT about what an PBI is, but other than that what do you expect. You don't learn coding in college, companies and devs need to train freshmen.

LLMs still cannot solve most logic problems. They generate language. It can be a programming language but obviously the scope is limited.

refurb• 2 years ago

It would be interesting if AI could be used to isolate bugs?

Have it run through all the branches of spaghetti code and isolate the problematic sections along with proposed changes.

scruple• 2 years ago

And that's to say nothing of the fact that the programs we build exist in an ecosystem of other programs, and sometimes you're lucky if you even have a copy of the source available... And that's without mentioning that it's all networked and you're dependent on other systems all the way down. So much of the work I do today is black box debugging where I can only make broad assumptions about the behavior of systems that are distant to my own based on their past performance, based on some notes someone took from a meeting 6 months ago, based on what the engineers on that team believe their systems are doing, etc...

A lot of times at work, support teams modify their systems and don't bother communicating what they're doing, they seem to rely on scream tests to figure out who is even using their services and for what purpose. It's a total fucking mess and I laugh whenever I hear someone suggest that "AI" is going to replace what I do when some massive % of what I do has fucking nothing to do with code at all.

jti107• 2 years ago

our company uses them LLM's extensively for coding and it was something that was optional the company paid for. we got training on how to use it and was told don't blame the LLM for shitty code, you're responsible for the work you push to the branch. most people ended up using it...our superstar dev uses it and he think he is way more productive with it.

Ferret7446• 2 years ago

Senior devs, no. Junior devs? Uh...

I've been seeing the same thing Steve Yegge has with AI. It could very well replace all of the grunt work in software engineering.

mattlutze• 2 years ago

If teams are hiring Junior developers to give them "grunt work," they're doing something wrong to begin with.

jerf2• 2 years ago

That probably felt very nice to say, but it's wrong. Everyone starts somewhere. Nobody can start out architecting the login infrastructure for the 250 applications at a major company, all of which already have their own login system and local considerations. Everybody starts out on some simple work. They have to. It's all they can do on day one. It's all I could do. It's all anyone can do.

I'm fairly worried that AI as it is developing now is going to be the most effective ladder pull in the world. While a new dev is grappling with "grunt work", they're learning. I'm working with them, explaining why this was done and why that was done, and why, yeah, that's busted, but actually not for the reason you think, and they're learning a lot more than just how to accomplish the specific task nominally laid out in the specific ticket they are working on. If they do the work by just asking an LLM for the answer and they actually get it, they may finish the task today... but what of their progress in five years? Ten? I'm very concerned that juniors leaning on AI will discover that in five years, the easy stuff is still easy, but the hard stuff is still as incomprehensible as it was five years ago.

mikeocool• 2 years ago

People keep saying this — but if I had to give a junior the level direction I give an AI bot everyday, beyond their first few weeks, that person would be on a pip pretty quickly.

Maybe I’m just fortunate, but most places I’ve worked the difference between jr and senior has been the scope they owned and the amount of time spent architecting/reviewing/mentoring vs heads down coding. Seniors are not just handing off grunt work to juniors.

lokar2• 2 years ago

The grunt work, not the hard work.

I do the hard work of software Eng away from my computer, trying to think clearly about the problem(s).

fzeroracer2• 2 years ago

Juniors are far more valuable to your organization than just grunt work. They offer fresh perspectives from an 'untainted' perspective. They might know about new technology or understand how customers interact with your system better than the senior engineer who has been behind the desk for so long that all they see is the bugs and issues. AI cannot do anything like this, it can only affirm what you ask of it and in the ways you ask it.

Of course many organizations don't like the whole 'training' people thing anymore so it's rough for junior devs out there. Good for my career since I can demand a premium since the pool of experienced engineers will only go down over time, but ultimately bad for software as a whole.

hindsightbias• 2 years ago

Given all these news articles are Ai generated I think they’re shouting at clouds now. The Ai knows how to generate clicks better than you do.

dbetteridge• 2 years ago

I personally like the idea of a "companion AI", it comes up a lot in video games and sci-fi stories (funnily enough ones where capitalism has been taken out back and shot.)

But the concept of a second brain that you can use to rubber duck concepts, decipher documentation and save your wrists from carpal tunnel when dealing with boilerplate is very useful.

With that all said, as a senior dev my day is only about 50-75% coding (on a good day) and the rest is often meetings/planning boards/bug triage/helping juniors and just plain translating product requirements into meaningful pieces of work for Devs, which I've not had much luck using an LLM to replace.

kgilpin• 2 years ago

I think people / the market have gotten a little too excited about something AI is actually pretty bad at - making changes to existing code (which is, after all, most of the code).

AI software devs don’t understand requirements well, and they don’t write code that confirms to established architecture practices. They will happily create redundant functions and routes and tables, in order to deliver “working code”.

So AI coding is bunk? No, it’s just that the primary value lies elsewhere than code generation. You can stuff A LOT of context into an LLM and ask it to explain how the system works. You can ask it to create a design that emulates existing patterns. You can feed it a diff of code and ask it to look for common problems and anti-patterns. You can ask it create custom diagrams, documentation and descriptions and it will do so quickly and accurately.

These are all use cases that assist with coding, yet don’t involve actually writing code. They make developers more knowledgeable and they assist with decision making and situational awareness. They reduce tedium and drudgery without turning developers into mindless “clickers of the Tab key”.

rohansood15• 2 years ago

Could not agree more. Plus, even for all of those tasks it takes a couple of iterations if not more.

j452• 2 years ago

I'm finding different tools can do this to different degrees of success.

Aider, for example, has been pretty decent at finding where to do what, as long as it's guided there in a certain way.

Still, it's far from perfect, but in more and more cases, doable.

aussieguy1234• 2 years ago

By the time AI can replace Human Devs, it'll also be able to replace almost all jobs that involve working on a computer.

That will trigger mass unemployment and our current economic system could collapse and need replacing with a new system, hopefully one where the profits of AI are shared around more equally.

As far as the jobs that could be automated away soon go, I'd say that being a developer is one of the least likely to be automated away in the short term and it may be one of the last to be automated.

intelVISA• 2 years ago

I prefer the Alternate Reality where businesses are now finally run as they should be: managers describing issues using technical terms that are fed into a machine to produce desired outputs.

Not to be confused with legacy arts like coding, or programming, which required Very Important project management frameworks... I guess we don't need PMs now either... or anyon-

kaushikc• 2 years ago

I don't care about writing code, all I care is that my work (non-tech) needs to be done. I code out of necessity to save more time, reduce labour and errors. LLM has made a non-programmer like myself and given me super powers, all I have to do is ask the right questions and I am directed to somewhat of a reasonable place to look for a solution and create ones. I can create CRUD apps, use api's and various stacks from multiple programming languages while referencing documentation to build and practice building pretty much what I desire which could have costed me a fortune and a lot of time from professional coders, some work I could not give to anyone else for maintaining confidentiality.

alphabettsy• 2 years ago

How is it different for you than using StackOverflow and GitHub?

tourmalinetaco• 2 years ago

LLM + search is far more powerful and intuitive than searching alone. I can actually explain what I want, what problems I have, and what I don’t want and it will (kinda) follow the directions laid out. It doesn’t always work, but it also hasn’t been too difficult to troubleshoot either. It absolutely cannot fix itself though, so if you can’t find the problem then you‘re very much stuck. All in all, it‘s different because it lets me delegate writing the code while I can focus on designing the program/problem solving, which I enjoy more anyway.

It doesn’t make non-programmers programmers, and it doesn’t make junior devs into senior devs, but at least for more basic problems/well-defined projects it can help turn a night of typing away into a night of development.

bruce511• 2 years ago

I've spent my career working in and on a low-code tool. For 30 years we've been doing what you describe; allowing non-coders to generate working applications that solve business problems.

It's similar to the way that Excel exposed non programmers to data manipulation.

Lots of people had very successful careers building software this way. They created a valuable product with mostly just domain knowledge and not programming knowledge.

But there are boundaries to this approach. Because they lack the fundamentals, they're (mostly) unable to understand what something is doing or how something works. So there's a skilled group providing assistance (paid) whenever they hit a boundary.

AI won't make programmers go away. But it will both expand the reach of "writing programs" to more people, and simultaneously replace "unskilled" programmers.

As more joe public writes code, more programmers will ultimately be needed to support those Joe's. And in some cases, just like spreadsheets, the help might be "we need to throw all this away and build it correctly."

mejutoco• 2 years ago

It sounds like you care about writing code and are learning to code. And that is a good thing.

hyperG• 2 years ago

I am in the same spot as a non-software engineer.

If you are using multiple programming languages though you really can't say you are a non-programmer.

I just got up and running with Julia in like two weeks when I didn't even know what Julia was 3 weeks ago.

On the other hand, I have the digital audio workstation Reaper installed. It is one thing for LLMs to write crud apps and data science scripts but to write a digital audio workstation(something I have 25 years of experience as a user), as a non-software engineer I wouldn't even know how to begin. Even if I could compile the code to play an audio file with a volume knob, the gulf is so huge. I wouldn't be able to add more than a few features before it would become totally unmanageable since I don't know what I am doing.

Much of what makes LLMs so impressive is we have such low expectations of modern software. They have broken the market for demo CRUD apps. Beyond that though I wouldn't trust anything cursor and I make.

tocs3• 2 years ago

I am curious to see how all this plays out. I am not a developer but have played some with chatGPT to write some simple python. It has done an OK job. When it works it is sometime easier than other cut and paste methods. That is a long way from replacing those that know what they are doing (even at a junior level).

On the other hand I remember hearing the old electrical engineers saying "you do not need a microcontroller for that. Just use a resistor and capacitor, or a 555".

As an aside, is my Firefox spell check worse? Are they getting ready to replace it with a LLM?

benreesman3• 2 years ago

LLM coding stuff is interesting. Sonnet is a class apart and I think I’ve upgraded it to unqualified “useful” on certain tasks.

But good luck trying to measure or even speculate about the impact that has on the market for software people.

The big picture is geopolitics (which is crazy right now), and then monetary policy (which has been all over the place for years and years), then politics and fiscal policy implications (which is crazy as hell right now), then this recent trend towards just brutally bare-fanged go-for-broke class warfare Thiel shit where FAANG is “tightening its belt” and smashing EPS simultaneously (boomeranging people for less RSUs and shit) with products that are now basically loathed but sticky like a stalker.

And all the quasi-consensus benchmarks of who’s even good are shot to hell: YC is funding Pear, Google can’t ship an LLM anyone wants, Amazon is just dark patterns and price fixing, they legendary shops just aren’t good anymore. Some of the best hackers I’ve met in decades are under-appreciated if they even have work, and “culture fit” (sociopath) knuckleheads are turning down windfalls.

It’s a fucking casino in 2024 and AI coding is a blip on this scale.

benreesman3• 2 years ago

@dang what is with sibling getting flagged dead for saying he agrees with (my) parent.

We’re both right but I understand that moderation has constraints that comments don’t.

Please either un-flag sibling or flag me.

benreesman3• 2 years ago

@dang I hate to push the point but I’m a real stickler for the vulnerable getting signal boosted rather than shouted down.

Maybe it is just legit flags that bury everything from Garry Tan tweeting “die motherfucker die” to elected officials, or Mike Seibel defrauding Autodesk with claims that SocialCam had novel technology before vesting in Thailand and ignoring emails for a year, or Emmett cashing out just in time for Twitch to fly into the side of a mountain, or for Altman to have been rescued from his abject incompetence via pg and Conway rigging up Green Dot. It goes on and on, and that’s for people who dismiss out of hand the allegations of sexual assault.

I’ve not only lived in the same building as Drew Houston, I’ve held his head when he was vomitting up mystery pills in Tahoe: and there is no evidence I’ve ever seen that any of these people are any good at anything other than “becoming powerful” to quote Paul.

I’ve just barely scratched the surface of the perversity, criminality, and sociopathy of the Culture Fit in 2024.

You’re one of the good guys, and I don’t envy the head that wears the crown on being both scrupulously fair and painting YC in a good light.

I know that it’s a small moderation team and that 3 humans don’t have the speed and clarity of an Azure IPv4 block.

But if you tell me that 3-5 is still the right number of people who can flag something dead while the community has grown by 7 orders of magnitude, no sir.

Flags are OP as fuck because you can count on the lie that comp.lang.lisp being hostile in bytes was worse than the contemporary fact that the Valley ends people’s livelihoods and smiles the whole time.

mattfrommars• 2 years ago

Neal Wu recently posted a video of him attending Meta Hack Coding challenge. This time around, Meta has two score board, one for all human and the other one, I believe is AI assisted. Shockingly, the leader board for human was significantly faster than leaderboard for folks who used AI.

We have a long way to go before AI surpasses human.

discordance• 2 years ago

A competition like that is for the top 1% and very difficult problems. For most people and most problems, AI is very helpful and speeds up the process.

As an example, I made a browser extension in 5 minutes this weekend using chatgpt by telling it what I wanted. I didn't know how to make a browser extension before that, so you could say AI surpassed this human quite easily.

krater234• 2 years ago

So AI is the next way to give people that don't know what they do new tools to generate and release shit where they don't know what it does to customers that don't know how much shit they got. I'm sure this will bring the mankind further...

jppope• 2 years ago

As expected, AI is great for stuff thats been done a million times. AI will slow you down for anything even slightly novel.

beeflet4• 2 years ago

Yeah I guess it helps for applications where you otherwise need to read a ton of documentation to get started. For web stuff it helps me find useful APIs sometimes, but other times it wastes my time with deceptively wrong answers.

The task of programming or engineering is generally one where you spend 80% of the time solving bugs in 20% of the code. LLMs are good for suggesting possible solutions based on existing things, but not for verifying solutions or doing any sort of real problem solving (they generally waste more time than they save because they give plausible but wrong answers).

If you just want to reproduce some boilerplate or example code it is great though because documentation tends to lack good examples, and LLMs have a ton of real code stored in their latent space. Sometimes I even just use it as a thesaurus, if I have something on the tip of my tounge.

rsynnott4• 2 years ago

Shockingly

Yeah, see I don't think that's at all shocking.

yashvg• 2 years ago

While I understand the desire to reassure developers, I think this perspective seriously underestimates the pace of progress in AI. Just 3-4 years ago, the idea of AI writing any functional code seemed far-fetched. Now they can handle many coding tasks competently.

The author lists specific tasks LLMs can't do today. But there's no fundamental reason they won't be able to in the near future. Domain expertise, understanding downstream effects, configuring CI pipelines - these are all learnable patterns. As models get larger, are trained on more diverse datasets, size of context window increases, and new architectures emerge, these capabilities will come online rapidly. The jump from GPT-3 to GPT-4 was substantial, and we should expect continued leaps.

This doesn't mean human developers will become obsolete overnight. But it does mean the nature of software development work is likely to change significantly. Lower-level coding tasks may be increasingly automated, shifting focus to higher-level design, architecture, and problem framing.

Rather than dismissing the potential impact, we should be preparing for a world where AI significantly augments or even replaces many current development tasks. This might involve focusing more on skills that complement AI capabilities or exploring new areas where human creativity and insight remain critical.

Aeolun2• 2 years ago

Maybe we should instead do those things when that time actually comes. Premature optimization and all that.

lurking_swe• 2 years ago

current AI is already stressing the power grid, and much of it will need to be redeveloped and improved just to keep pushing the limits of LLM’s. Power is the limiting factor with scaling here, so i’m rather unconvinced with your hypothesis. The improvements in the last 2-3 years are in no way indicative of the next 2-3 years.

I agree with your sentiment by the way, developers should find ways to use LLM’s to improve their development process. But the drama is getting old.

zubairq3• 2 years ago

I “wish” AI would replace me so that I could do other stuff than coding

93po5• 2 years ago

ive loved the past two weeks of working on a project where cursor is doing 98% of the coding, i have a sweet project at 20x the normal speed i could do it

zubairq3• 2 years ago

Thanks. I've tried Cursor, but probably more useful on smaller codebases I found and for simple code changes. Can you share the github repo that you use it on, or give me an indication of the codebase size that you use it on?

Aeolun2• 2 years ago

I'm a developer myself (2k+ commits 3yrs in a row)

What a completely weird thing to identify yourself by number of commits. I can make 5k+ commits in 3 days. Does that make me a better developer?

purple-leafy• 2 years ago

Yeah, low number of commits too if they are trying to brag. I did a PR the other day that was ~50 commits, and I only took a few hours.

At that speed working 40 days I’m already blasting this guys number of commits every year lol

Stupid metric, stupid post

lispisok• 2 years ago

I could see it as being a metric like lines of code written. The days I make more commits is usually because I coding poorly and having to fix my bugs and fix missed edge cases. Fewer well thought out tested changes are better

noobermin• 2 years ago

All I'll say is if you for the past few years were chiming in with how this will change everything only to join the rest of us skeptics in the last year or so, you don't get to turn around and pretend you were right all along about the limitations, that AGI was never going to happen with LLMs only, etc, etc. I just wish some of you , who let's be real, just jumped off the crypto bandwagon a few years ago, would do some introspection first.

Btw, if you do agree with OP about development, extend that to art, writing, etc. The same kind of deep domain specific knowledge is required in almost all things that humans do.

segmondy• 2 years ago

Well, it's not a matter of if AI will replace human devs, it will.

The question is WHEN and HOW many devs?

Not all devs will be replaced, but how much will be replaced and when will it start?

If you take your AI and your prompt it to build an app, are you not a dev? I believe the same way we have redefined AI and AGI numerous times, we will redefine what it means to be a developer.

owenpalmer• 2 years ago

LLMs cannot update the SDK code to V2 response types in the first place

Why not? Isn't it just a matter of adding the new V2 response type to the context window and asking it to update those types in another file?

LLMs cannot successfully configure a CI action such that I won't have to publish myself

Why can't it do this? I'd like to see your prompt.

LLMs do not understand the downstream effect of fixing the problem will be updating yournextstore's code

I suppose I don't understand the specifics enough to analyze why this isn't doable with an LLM. However, it seems like the theme is that code almost always relies on other code, and if one part breaks, it's hard to get it agreeing again. To solve this, you could have the LLM agent routinely add changelogs to the context window when making maintenance changes.

If you can't get an LLM to do something, you probably aren't thinking creatively enough.

leshenka• 2 years ago

No amount of "please don't use features that do not exist" will stop LLM from making stuff up

kkfx• 2 years ago

Anyone who have ever tried to generate code via LLMs know that. The main hype point though it's selling something that can't work but people believe it do to hide a very simple phenomenon: after DECADES of technical possibility and opportunity IT automation timidly became a bit spread, so finally people have discovered they do not need to physically go to a bank to operate with it, since web-banking is here since decades, they discover they do not need an insurance broker for 99% of insurances instead they choose some on-line offer comparison portal, they do not even need an office to do office works since without paper we can WFH.

This obviously means many jobs are not needed anymore. Much of ETL, office frontline etc for instance. But if you tell anyone this is not new, it's decade old stuff you are clearly a bad manager because you could makes things much better much before, of you simply state that most people are ignorant of IT and that's a dramatic problem for the society because if we know IT we can create new jobs, if not we can only dismiss many and so on. So "hej, it's AI!" [and obviously that's new eh]...

Aside the office use of LLM could led to interesting nightmares in infosec terms, so brace yourself...

ookblah• 2 years ago

currently AI focus is very narrow. I treat it like a junior dev where it can repeat structure it's given and then I basically "approve" it or I use it like a super robust autocomplete where it knows what I'm writing next like 80% of the time. also good for bouncing ideas to see if there's something someone else had done similarly.

there is no way in hell I would tell it to "generate X feature" or "fix X bug" wholesale. half the time it is just dead wrong in it's approach and it's more time consuming to grok what it did exactly and then either 1) suggest an alternate approach and 2) fix what it did wrong.

like i get it, in the end if it generates the right code and it works then most people will call it a day. but longevity is in being able to understand and maintain your codebase and at this point i can't trust it to do that. it some sense it reminds me of (and still reminds me of) those wysiwyg site builders where sure, your output might be fine for 90% of those cases like a simple landing page but the code underneath is a big pile of shit. good luck if you need to break out of that box at any point.

i guess all to say AI will start replacing low level static site facing stuff. for instance if you're selling cookie cutter WP themes i'd be worried at some point. for everything else it's a non-factor right now.

ristos• 2 years ago

I feel like the hype around LLMs replacing dev work is similar to what happened years ago when they were saying the same thing about scripting languages and WYSIWYG editors like Dreamweaver...

What ended up happening in practice is just that time was saved on grunt work, and really engineers just ended up working on composing and debugging higher level components, optimization, or low level work.

AI won't replace human devs anytime soon

Related Stories