15 ms·

For those less confident: U.S. (and German) automakers were absolutely sure that the Japanese would never be able to touch them. Then Koreans. Now Chinese. Now

For those less confident: U.S. (and German) automakers were absolutely sure that the Japanese would never be able to touch them. Then Koreans. Now Chinese. Now there are tariffs and more coming to save jobs. Betting against AI (or increasing automation, really) is a bet against not against robots, but against human ingenuity. Humans are the ones making progress, and we can work with toothpicks as levers. LLM's are our current building blocks, and people are doing crazy things with them. I've got almost 30 years experience but I'm a bit rusty in e.g. web. But I've used LLM's to build maybe 10 apps that I had no business building, from one-off kids games to learn math, to building a soccer team generator that uses Google's OR tools to optimise across various axes, to spinning up four different test apps with Replit's agent to try multiple approaches to a task I'm working on. All the while skilling up in React and friends. I don't really have time for those side-quests but LLM's make them possible. Easy, even. The amount of time and energy I'd need pre-LLM's to do this makes this a million miles away from "a waste of time". And even if LLM's get no better, we're good at finding the parts that work well and using that as leverage. I'm using it to build and check datasets, because it's really good at extraction. I can throw a human in the loop, but in a startup setting this is 80/20 and that's enough. When I need enterprise level code, I brainstorm 10 approaches with it and then take the reins. How is this not valuable?

Timber-6539 2y ago
This is no different from creating a to-do app with an LLM and proclaiming all developers are up for replacement. Demos are not what makes LLMs good, let alone useful.
apwell23 2y ago
> 've got almost 30 years experience but I'm a bit rusty in e.g. web. But I've used LLM's to build maybe 10 apps that I had no business building, from one-off kids games to learn math, Yea i built bunch of apps when RoR blog demo came out like 2 decades ago. So what?
nidnogg 2y ago
As a senior-ish programmer who struggled a lot with algorithmic thinking in college, it's really awe-inspiring. Truly hit the nail on the head there. We HAD no business with these side-quests, but now? They're all ripe for the taking, really.
cadamsau 2y ago
One hundred per cent this. LLM pair programming is unbelievably fun, satisfying, and productive. Why type out the code when you can instead watch it being typed while thinking of and typing out/speaking the next thing you want. For those who enjoy typing, you could try to get a job dictating letters for lawyers, but something tells me that’s on the way out too.
- not_kurt_godel 2y ago
  My experience is it's been unbelievably fun until I actually try to run what it writes and/or add some non-trivial functionality. At that point it becomes unbelievably un-fun and frustrating as the LLM insists on producing code that doesn't give the outputs it says it does.
- nidnogg 2y ago
  Yeah I had my fair share of pride around typing super fast back in college, but the algorithms were super annoying to think through. Nowadays I get wayyy more of a kick typing the most efficient Lego prompts in Claude.
- wyclif 2y ago
  What do you prefer to use for LLM pair programming?
  - cadamsau 2y ago
    Claude and pasting code back and forth for simple things. But I’d like to try Claude with a few MCP servers so it can directly modify a repo. But lately Cursor. It’s just so effortless.
  - richardw 2y ago
    Claude 70%. ChatGPT o1 for anything that needs more iterations, Cursor for local autocomplete, tested Gemini for concepts and it seemed solid. Replit when I want it to generate everything, from setting up a DB etc for any quick projects. But it’s a bit annoying and drives into a ditch a lot. I honestly have to keep a tight rein on them all, so I usually ask for concepts first with no code, and need to iterate or start again a few times to get what I need. Get clear instructions, then generate. Drag in context, tight reins on changes I want. Make minor changes rather than wholesale. Tricks I use. “Do you have any questions?” And “tell me what you want to do first.” Trigger it into the right latent space first, get the right neurons firing. Also “how else could I do this”. It’ll sometimes choose bad algo’s so you need to know your DSA, and it loves to overcomplicate. Tight reins :) Claude’s personality is great. Just wants to help. All work best on common languages and libraries. Edge cases or new versions get them confused. But you can paste in a new api and it’ll code against that perfectly. I also use the API’s a lot, from cheap to pricy depending on task. Lots of data extraction, classifying. I got a (pricier model) data cleaner working on other data generated by a cheaper model, asking it to check eg 20 rows in each batch for consistency. Did a great job.
dustingetz 2y ago
quantum computers still can’t factor any number larger than 21
jayd16 2y ago
"Other people were wrong about something else so that invalidates your argument" Why are half the replies like this?
- richardw 2y ago
  Because what is shared is overconfidence in the face of a system that has humble beginnings but many people trying to improve it. People have a natural bias against massive changes, and assume the status quo is fixed. I’m open to all possibilities. There might be a near term blocker to improvement. There might be an impending architectural change that achieves AGI. Strong opinions for one or the other with no extremely robust proof are a mistake.
- pdimitar 2y ago
  The cognitive dissonance of seemingly educated people defending the LLMs when it comes to writing code is my top mystery for the entirety of 2024. Call me if you find a good reason. I still have not.
  - sigmarule 2y ago
    Genuine question: how hard have you tried to find a good reason?
    - pdimitar 2y ago
      Obviously not very hard. And looking at the blatant and unproven claims on HN gave me the view that the proponents are not interested in giving proof; they simply want to silence anyone who disagrees LLMs are useful for programming.
  - jayd16 2y ago
    The short answer is they're trying to sell it.
    - richardw 2y ago
      And it’s the short wrong answer. I get absolutely zero benefit if you or anyone else uses it. This is not for you or I, it’s for young people who don’t know what to do. I’m a person who uses a few LLM’s, and I do so because I see the risks and want to participate rather than have change imposed on me.
pdimitar 2y ago
In other words, you have built exactly zero commercial-grade applications that us the working programmers work on building every day. LLMs are good for playing with stuff, yes, and that has been implied by your parent commenter as well I think. But when you have to scale the work, then the code has to be easy to read, easy to extend, easy to test, have extensive test coverage, have proper dependency injection / be easy to mock the 3rd party dependencies, be easy to configure so it can be deployed in every cloud provider (i.e. by using env vars and config files for modifying its behavior)... and even more important traits. LLMs don't write code like that. Many people have tried with many prompts. It's mostly good for just generating stuff once and maybe do little modifications while convincingly arguing the code has no bugs (and it has). You seem to confuse one-off projects that have zero to little need for maintenance for actual commercial programming, perhaps? Your analogy with the automakers seems puzzlingly irrelevant to the discussion at hand, and very far from transferable to it. Also I am 100% convinced nobody is going to protect the programmers; business people trying to replace one guy like myself with 3 Indians has been a reality for 10 years at this point, and amusingly they keep failing and never learning their lesson. /off-topic Like your parent commenter, if LLMs get on my level, I'd be _happy_ to retire. I don't have a super vested interest in commercial programming, in fact I became as good at it in the last several years because I started hating it and wanted to get all my tasks done with minimal time expended; so I am quite confident in my assessment that LLMs are barely at the level of a diligent but not very good junior dev at the moment, and have been for the last at least 6 months. Your take is rather romantic.
- Aeolun 2y ago
  Very defensive. I love it anyhow. Sure, it generates shit code, but if you ask it it’ll gladly tell you all the ways it can be improved. And then actually do so. It’s not perfect. I spent a few hours yesterday pulling it’s massive blobby component apart by hand. But on the plus side, I didn’t have to write the whole thing. Just do a bunch of copy paste operations. I kinda like having a junior dev to do all the typing for me, and to give me feedback when I can’t think of a good way to proceed.
  - olivermuty 2y ago

Timber-6539 2y ago

This is no different from creating a to-do app with an LLM and proclaiming all developers are up for replacement. Demos are not what makes LLMs good, let alone useful.

apwell23 2y ago

> 've got almost 30 years experience but I'm a bit rusty in e.g. web. But I've used LLM's to build maybe 10 apps that I had no business building, from one-off kids games to learn math, Yea i built bunch of apps when RoR blog demo came out like 2 decades ago. So what?

nidnogg 2y ago

As a senior-ish programmer who struggled a lot with algorithmic thinking in college, it's really awe-inspiring. Truly hit the nail on the head there. We HAD no business with these side-quests, but now? They're all ripe for the taking, really.

cadamsau 2y ago

One hundred per cent this. LLM pair programming is unbelievably fun, satisfying, and productive. Why type out the code when you can instead watch it being typed while thinking of and typing out/speaking the next thing you want. For those who enjoy typing, you could try to get a job dictating letters for lawyers, but something tells me that’s on the way out too.

not_kurt_godel 2y ago
My experience is it's been unbelievably fun until I actually try to run what it writes and/or add some non-trivial functionality. At that point it becomes unbelievably un-fun and frustrating as the LLM insists on producing code that doesn't give the outputs it says it does.
nidnogg 2y ago
Yeah I had my fair share of pride around typing super fast back in college, but the algorithms were super annoying to think through. Nowadays I get wayyy more of a kick typing the most efficient Lego prompts in Claude.
wyclif 2y ago
What do you prefer to use for LLM pair programming?
- cadamsau 2y ago
  Claude and pasting code back and forth for simple things. But I’d like to try Claude with a few MCP servers so it can directly modify a repo. But lately Cursor. It’s just so effortless.
- richardw 2y ago
  Claude 70%. ChatGPT o1 for anything that needs more iterations, Cursor for local autocomplete, tested Gemini for concepts and it seemed solid. Replit when I want it to generate everything, from setting up a DB etc for any quick projects. But it’s a bit annoying and drives into a ditch a lot. I honestly have to keep a tight rein on them all, so I usually ask for concepts first with no code, and need to iterate or start again a few times to get what I need. Get clear instructions, then generate. Drag in context, tight reins on changes I want. Make minor changes rather than wholesale. Tricks I use. “Do you have any questions?” And “tell me what you want to do first.” Trigger it into the right latent space first, get the right neurons firing. Also “how else could I do this”. It’ll sometimes choose bad algo’s so you need to know your DSA, and it loves to overcomplicate. Tight reins :) Claude’s personality is great. Just wants to help. All work best on common languages and libraries. Edge cases or new versions get them confused. But you can paste in a new api and it’ll code against that perfectly. I also use the API’s a lot, from cheap to pricy depending on task. Lots of data extraction, classifying. I got a (pricier model) data cleaner working on other data generated by a cheaper model, asking it to check eg 20 rows in each batch for consistency. Did a great job.

dustingetz 2y ago

quantum computers still can’t factor any number larger than 21

jayd16 2y ago

"Other people were wrong about something else so that invalidates your argument" Why are half the replies like this?

richardw 2y ago
Because what is shared is overconfidence in the face of a system that has humble beginnings but many people trying to improve it. People have a natural bias against massive changes, and assume the status quo is fixed. I’m open to all possibilities. There might be a near term blocker to improvement. There might be an impending architectural change that achieves AGI. Strong opinions for one or the other with no extremely robust proof are a mistake.
pdimitar 2y ago
The cognitive dissonance of seemingly educated people defending the LLMs when it comes to writing code is my top mystery for the entirety of 2024. Call me if you find a good reason. I still have not.
- sigmarule 2y ago
  Genuine question: how hard have you tried to find a good reason?
  - pdimitar 2y ago
    Obviously not very hard. And looking at the blatant and unproven claims on HN gave me the view that the proponents are not interested in giving proof; they simply want to silence anyone who disagrees LLMs are useful for programming.
- jayd16 2y ago
  The short answer is they're trying to sell it.
  - richardw 2y ago
    And it’s the short wrong answer. I get absolutely zero benefit if you or anyone else uses it. This is not for you or I, it’s for young people who don’t know what to do. I’m a person who uses a few LLM’s, and I do so because I see the risks and want to participate rather than have change imposed on me.

pdimitar 2y ago

In other words, you have built exactly zero commercial-grade applications that us the working programmers work on building every day. LLMs are good for playing with stuff, yes, and that has been implied by your parent commenter as well I think. But when you have to scale the work, then the code has to be easy to read, easy to extend, easy to test, have extensive test coverage, have proper dependency injection / be easy to mock the 3rd party dependencies, be easy to configure so it can be deployed in every cloud provider (i.e. by using env vars and config files for modifying its behavior)... and even more important traits. LLMs don't write code like that. Many people have tried with many prompts. It's mostly good for just generating stuff once and maybe do little modifications while convincingly arguing the code has no bugs (and it has). You seem to confuse one-off projects that have zero to little need for maintenance for actual commercial programming, perhaps? Your analogy with the automakers seems puzzlingly irrelevant to the discussion at hand, and very far from transferable to it. Also I am 100% convinced nobody is going to protect the programmers; business people trying to replace one guy like myself with 3 Indians has been a reality for 10 years at this point, and amusingly they keep failing and never learning their lesson. /off-topic Like your parent commenter, if LLMs get on my level, I'd be _happy_ to retire. I don't have a super vested interest in commercial programming, in fact I became as good at it in the last several years because I started hating it and wanted to get all my tasks done with minimal time expended; so I am quite confident in my assessment that LLMs are barely at the level of a diligent but not very good junior dev at the moment, and have been for the last at least 6 months. Your take is rather romantic.

Aeolun 2y ago
Very defensive. I love it anyhow. Sure, it generates shit code, but if you ask it it’ll gladly tell you all the ways it can be improved. And then actually do so. It’s not perfect. I spent a few hours yesterday pulling it’s massive blobby component apart by hand. But on the plus side, I didn’t have to write the whole thing. Just do a bunch of copy paste operations. I kinda like having a junior dev to do all the typing for me, and to give me feedback when I can’t think of a good way to proceed.
- olivermuty 2y ago