Thank you for your thoughtful take on this, Ryan! I really appreciate the hype-free, honest, non-sponsored look at these tools from an experienced engineer that I trust.
I tried it for a small personal project after reading this post. ~75-85% of the time it will make great suggestions and let me blow through accepting them and making a bunch of edits to my code. The other 15-25% of the time it gives me irredeemably bad suggestions, or straight-up hallucinated nonsense. On balance, that's still a huge time-saver, since so much of web development is just mundane repetitive boilerplate.
Overall, would keep using, especially in the context of small side projects where I don't care about my code being hoovered up into the cloud.
Thanks again for the write-up and the push to check this out!
Glad to hear its helping you with personal projects too Mike.
Had a similar experience with the terrible suggestions some of the time. Overall a huge time saver. Glad to not have to learn random frameworks to get little things done anymore.
I am an AI engineer and all these llms looks great, fancy and promising until we have clean and useful data to use and train them. If languages will not update anymore then we will have them useful for long time. Otherwisr if most devs become lazy and depend on these llms more then there wont be solutions on the internet websites like stackoverflow or quality code on github to use for training these llms. Which would result into these becoming obsolete sooner or later.
And languages will keep updating and changing. You have seen the era of angularjs and then transition to angular 2 and then there was react. Now we got nextjs. The browsers become efficient the space and memory is getting better so the languages are also evolving. Code related llms cannot stay same like other use cases.
Its simple. Python 3.4 was much different (without typings, dependencies etc) from current python 3.11. If llm is trained with information about 3.4 then it wont help for 3.11. There is no magic going on inside the llm. Its summarized data trained from original quality data. If more people will rely on the llms then there would be less data to train with. Which ultimately result into these llms become less useful
llm have better answer than stackoverflow.... and it has no mods which some times could be really annoying.... if i'm debugging late at night 10 pm do i filtering out none sense SOF answer or my sonnet? it is not true that people will not go to stack overflow and thus the website is completely dead, i go SOF after i dont get anything from my sonnet, SOF gonna have a smaller size of answer but the content will be focus on new questions and greater quality answers...
Seems like it could boost efficiency for an experienced developer by a marginal amount. But for juniors, it might spell disaster if they accept the kind of boilerplate code produced by the LLM too readily. Juniors won't have the experience or battle-tested knowledge of what bad code smells, anti-patterns, and downright broken syntax or hallucinated libraries to watch out for. How do we justify the rapid proliferation of tools that help advanced developers but may totally mislead beginners or even eradicate those junior roles?
To me, LLMs are just a faster way of writing simple code. No need to throw away code review practices that help junior engineers learn from more senior ones.
Why do people always assume that software is developed in full isolation, by a lone person sitting by him/herself in front of the computer? That's so 20th century. We're in 2024 today.
Appreciate the more technical read, Ryan! Definitely happy to be an early tester of your tool as well. I used Cursor to revamp my personal website, but I like your idea of creating an editing tool a lot better 😂
Agree. I also found claude 3.5 sonnet amazing for making my writing clearer
Thank you for your thoughtful take on this, Ryan! I really appreciate the hype-free, honest, non-sponsored look at these tools from an experienced engineer that I trust.
I tried it for a small personal project after reading this post. ~75-85% of the time it will make great suggestions and let me blow through accepting them and making a bunch of edits to my code. The other 15-25% of the time it gives me irredeemably bad suggestions, or straight-up hallucinated nonsense. On balance, that's still a huge time-saver, since so much of web development is just mundane repetitive boilerplate.
Overall, would keep using, especially in the context of small side projects where I don't care about my code being hoovered up into the cloud.
Thanks again for the write-up and the push to check this out!
Glad to hear its helping you with personal projects too Mike.
Had a similar experience with the terrible suggestions some of the time. Overall a huge time saver. Glad to not have to learn random frameworks to get little things done anymore.
Nice read, curious to see how you think LLMs will learn "newer" bugs if people stop going to sites and use LLMs.
Who's going to create?
Perhaps LLMs can generate their own data at some point, I have no idea though to be honest haha
I am an AI engineer and all these llms looks great, fancy and promising until we have clean and useful data to use and train them. If languages will not update anymore then we will have them useful for long time. Otherwisr if most devs become lazy and depend on these llms more then there wont be solutions on the internet websites like stackoverflow or quality code on github to use for training these llms. Which would result into these becoming obsolete sooner or later.
I hear that argument a lot. I don't fully buy it since it assumes we can't generate synthetic data for future models
Not convinced humans are needed forever, but I'm not a subject matter expert so I don't feel strongly
And languages will keep updating and changing. You have seen the era of angularjs and then transition to angular 2 and then there was react. Now we got nextjs. The browsers become efficient the space and memory is getting better so the languages are also evolving. Code related llms cannot stay same like other use cases.
Its simple. Python 3.4 was much different (without typings, dependencies etc) from current python 3.11. If llm is trained with information about 3.4 then it wont help for 3.11. There is no magic going on inside the llm. Its summarized data trained from original quality data. If more people will rely on the llms then there would be less data to train with. Which ultimately result into these llms become less useful
llm have better answer than stackoverflow.... and it has no mods which some times could be really annoying.... if i'm debugging late at night 10 pm do i filtering out none sense SOF answer or my sonnet? it is not true that people will not go to stack overflow and thus the website is completely dead, i go SOF after i dont get anything from my sonnet, SOF gonna have a smaller size of answer but the content will be focus on new questions and greater quality answers...
you should change “A new era of No coding” instead
I am not from IT domain, your article gave me deep understanding of new era of AI in Coding. It would take more time to fully replace human minds.
Seems like it could boost efficiency for an experienced developer by a marginal amount. But for juniors, it might spell disaster if they accept the kind of boilerplate code produced by the LLM too readily. Juniors won't have the experience or battle-tested knowledge of what bad code smells, anti-patterns, and downright broken syntax or hallucinated libraries to watch out for. How do we justify the rapid proliferation of tools that help advanced developers but may totally mislead beginners or even eradicate those junior roles?
To me, LLMs are just a faster way of writing simple code. No need to throw away code review practices that help junior engineers learn from more senior ones.
From our point of view, sure. But what does the CTO/CEO say? What would be the best outcome from a short-term investor value perspective?
Why do people always assume that software is developed in full isolation, by a lone person sitting by him/herself in front of the computer? That's so 20th century. We're in 2024 today.
Now what’s the side project? Share it with us 😊
I am building a writing assistant which fits my editing needs. I will share the tool when I get it more polished!
100% agree here, recently switched to cursor and there's no way back.
What did you build in a weekend?
I built a writing assistant I've always wanted, it's a little chrome extension that automates my editing workflow for writing
I'm going to polish it a bit more then share it around and see what people think
Appreciate the more technical read, Ryan! Definitely happy to be an early tester of your tool as well. I used Cursor to revamp my personal website, but I like your idea of creating an editing tool a lot better 😂
Thanks Jordan!
I have been neglecting my personal website for some time that's a great idea, I think I'll clean revamp mine too since Cursor makes it so easy