I am a 14 y experienced sw engineer and 6 years experienced entrepreneur. Today, ChatGPT 4 does 80% of the coding for me.
Today it wrote two blog posts and generated the artwork and SEO tags for it.
Recently ChatGPT put together a thorough research on what’s the best available solar power station on the market for my niche application. It saved me a day of research.
In one year I will be a better promoter, will have better tools, and the AI will be quicker and more powerful.
There’s absolutely no way back from this.
BTW I pay for copilot too, but it helps very little compared to my blatantly suboptimal ChatGPT copy + paste + prompt workflow.
I have maybe 40 years experience of software development in every language conceivable from assembler to Visual Basic.
ChatGPT4 does 50% of my coding for me. And 90% of finding my stupid errors.
There was a function I had a couple of days ago that had some gnarly SQL in it and it now needed to return two variables instead of one. I just looked at the code and sighed because of the effort I knew it would take to refactor the SQL and the surrounding code.
Then I pasted it into GPT and wrote half a sentence on what the changes were and it spat out 100% perfect version of exactly what I needed.
One of the only downsides is that GPT's data is out-of-date, especially as I'm working under some tools that only really came out since its cut-off. So I often have to adapt its output.
I have hardcore ADHD and some very mild dyslexia. It has prevented me from being able to really code my entire life. I've always been able to understand small scripts and that has helped me piece stuff together my entire career. I just always felt like I was hitting a brick wall when I would try and create stuff myself. I could learn it, go through all of the exorcises to successfully complete guided projects, but when it came to coding from scratch I have severe recall issues and can not produce anything to save my life.
ChatGPT has opened up an entirely new world to me. I have (somewhat) successfully created many working applications, including a Flask site that can summarize news articles and websites and also create an audio version of the summarized text. I've created a Flask site that displays system usage for all of my home lab devices. I've written PowerShell and Bash scripts for work and personal use. It's become the workaround I needed for my "disabilities" and amplified my abilities to another level.
I use Copilot in VS Code and I was highly sceptical of it at first. It's true that it will often make mistakes if you ask it for complex things, but for me the real value is in auto-completing all the low-value tasks, like writing docstrings and refactoring existing code. I do use it to write new code from time to time, but I always validate the generated code and never allow it to generate more than a few lines at a time. In saying that, it can be fun to write a comment explaining the code I'm about to write, and then letting Copilot generate the code itself based on the comment. It's very rare that Copilot will get it fundamentally wrong - it's usually spot-on or mostly correct.
I also use ChatGPT for higher-level questions - like figuring out an approach to solve a problem, or asking it for suggestions on how to refactor a file.
I pay for both and get tremendous value from both.
I agree. It's really like having a buddy next to me, helping me out. It motivates me too, because there are some tasks that I am so demotivated to work on, but GPT helps me through them.
First time I'm hearing about Cursor.sh, this looks really powerful, does anyone else have similar experiences to share? Is it a notch above VS code with the Github Copilot plugin?
Copilot is like supercharged autocomplete that actually works. Writes all your boilerplate and finishes most of your lines with what you intended to write. Definitely worth its money. Using an editor without it feels like returning to the stone age.
But, how did this post end up on the front page with 3 upvotes?
I use Bing AI chat a lot in a simar way i used to google for stackoverflow answers. Only now i feed it with real function names, versions etc in the question itself. It will nicely provide an answer that already has those details filled in and is adapted to my particular context. And then when something is not clear about why that would even work, i keep on asking questions and i learn in more detail how it works. Sometimes i reply with 'but why not use...' and then it replies with 'actually yes...' and improves the answer. Nice partner to have!
Most of the time it’s like I spend 90% of the time trying to understand context of preexisting application and requirements, that’s purely reading and not writing
Once I start writing, it’s normally straightforward what’s needed. What difference does it make if it takes 10 minutes to write something or 2 minutes to write it? Especially when most nontrivial problems can take several hours to grok
And no, I don’t trust a model to “grok” the code for me, I’d much rather suffer through it. And introducing another tool isn’t worth 8 minutes of time savings to me if those 8 minutes are relatively enjoyable
LLMs are not suitable for some types of work but they seem suitable for others. ChatGPT hasn't changed my workflow drastically but I do use it at least once a day for one task or another. For example I just created this PowerShell script I would normally wouldn't bother to do to change some settings, send some automated emails etc.
The reading part is what I find really valuable about these tools. As I'm trying to understand a new code base I can ask questions and get excellent feedback. It is a true accelerator for understanding code, especially when unfamiliar frameworks are being used.
As a developer who has been coding for last 20 years, the best use of Copilot is to hit the tab. It is not that what to write next is problem, its faster and easier, specially in verbose languages like Java.
As others point out, generating docs is another great example. In one obscure case, Copilot was able to help fix a bug when interfacing between Golang/CGO/Cocoa framework - I had deleted the code block to rethink and it just displayed the suggestion. I just tried it by luck and it worked.
Both are great. Actually the free version of ChatGPT, 3.5, is good enough for most cases. Note that ChatGPT is 'offline' though, you need to copy-paste code back into your IDE.
Copilot is one notch above because it integrates directly into your IDE. If you follow coding channels on Youtube, you will notice a lot of creators use it for auto-fill, that's where I first saw the power of Copilot and am using it now.
I've used this one as well, but in terms of functionality, isn't it similar to GitHub Copilot? I find that GitHub Copilot in general works better and seems to understand context a bit better than the ChatGPT + VS code plugin.
I'm a Development Director and I manage a team of 74. My job is consists of a lot of meetings and other interruptions.
ChatGPT and other AI tools have lead to me increasing both my writing output increasing and having enough time to write code between those. That said, I'm not writing new features, more just refactoring and fixing up shit stuff.
In addition to code generation, our team has found the new AI code review tools to be quite useful as well. We use CodeRabbit and we keep finding issues/improvements in every other PR.
Not really as too much time is wasted on broken or incomplete results. But i did notice it gives the impression of helping junior developers or those with lengthy but weak experience. These tools almost always produce immediately visibly broken code or are useful for trivialities. Google does it better when needed.
Not particularly. I value predictivity over "smarts" when it comes to productivity tools. For me, using the already existing autocomplete, that I know the outcome of, is faster than understanding (and subsequently fixing) copilot’s suggestions.
A colleague types no more than 30 words per minute, and is fawning over copilot. There’s probably a correlation.
Today it wrote two blog posts and generated the artwork and SEO tags for it.
Recently ChatGPT put together a thorough research on what’s the best available solar power station on the market for my niche application. It saved me a day of research.
In one year I will be a better promoter, will have better tools, and the AI will be quicker and more powerful.
There’s absolutely no way back from this.
BTW I pay for copilot too, but it helps very little compared to my blatantly suboptimal ChatGPT copy + paste + prompt workflow.