1–9 of 9 posts
Is anyone thinking about this?
I've tried getting chatgpt and others to write code for me but it just seems to get confused about most everything once you get past the "write a function that does X" and X is some well known thing. Mostly I use it to suggest me a bunch of different ways to solve a problem so I can do research while I think about it. And I've been using it on an R library I'm developing and it ends up being a crutch. It's like follo…
What if it wasn't broken or confused and it worked immediately...like for example you don't just generate the code as output, instead it is immediately deployed and usable?
you should check out this: https://github.com/AntonOsika/gpt-engineer https://github.com/e2b-dev/awesome-ai-agents
For example, you can easily pay $10000 to a trendy developer to build a chatbot for your data, or you can use one of dozens of web applications to do the same thing for less than $100.
Vercel now has a system for generating React UIs which could literally replace hiring a second developer in many projects (you would still need a human programmer for some things I think).
I have built web apps that can iterate on programs or shell scripts and directly launch servers in VMs for you.
The trick is getting this stuff to be robust enough in the face of infinite potential requirements. When you narrow the scope down and add a bit of an error correcting loop, GPT-4 can absolutely create working applications.
I am working on an agent framework and hosting system that will have agent that can build agents or apps.
This is going to be a huge trend in the next couple of years. It will put pressure on the software development and start to take more and more low-budget projects away from programmers. If course this is already happening. For example, you can easily pay $10000 to a trendy developer to build a chatbot for your data, or you can use one of dozens of web applications to do the same thing for less than $100. Vercel now h…
I've tried getting chatgpt and others to write code for me but it just seems to get confused about most everything once you get past the "write a function that does X" and X is some well known thing. Mostly I use it to suggest me a bunch of different ways to solve a problem so I can do research while I think about it. And I've been using it on an R library I'm developing and it ends up being a crutch. It's like follo…
Thanks for sharing! I have heard similar feedback from others. What if it wasn't broken or confused and it worked immediately...like for example you don't just generate the code as output, instead it is immediately deployed and usable?