HN Explorer

Show HN: GPT Engineer – platform for devs to tinker with AI programming tools (antonosika.substack.com)

6 points by antonoo ↗ HN
Hello Hacker News community,

Wanted to share a project I’ve been working on during my spare time.

GPT Engineer’s mission: Create a flexible tool that anyone can experiment with to understand the current limits and future potential of natural language-prompted code generation

I believe it's key for us devs to engage in how building software can, and will, change.

Demo: https://twitter.com/antonosika/status/1667641038104674306

The project had a few broken commits and been a bit unreliable last week which felt quite bad, but it is now quickly getting better. The focus right now is to set up a benchmark for following the progress and clearly knowing current limitations. I'll share the benchmark results soon.

Appreciate any suggestions, experiences, or ideas on this project!

9 comments

[ 4.2 ms ] story [ 33.4 ms ] thread
skilled
I submitted this also but I haven't had a chance to use it yet.

Have you built anything useful with it beyond the very simple demo you showcased? Any issues you have encountered?

antonoo
We set up the benchmark recently and it does fail on quite a few of the more full stack example prompts we gave it yes.

There are clear remediations though so we want to continue building!

fantalamera
I used this when setting up a new codebase. It used and pointed out different Python libraries I didn't know existed. Pretty helpful
seanthemon
What's the token usage like? Will we run into high usage charges? Cool project!
antonoo
Yeah it's using GPT4 per default – I think it uses ~2k tokens for large projects!
dybekk
Do you think such tool can replace/reduce the need for software engineers. How do you feel about it?
antonoo
Definitely.

I think most of us will all be happy and only work out of passion in a few years, and I'm excited to learn new skills:)

j-a-a-p
Great initiative! Playing with it at the moment.

In your mission: Create a flexible tool that anyone can experiment with to understand the current limits and future potential of natural language-prompted code generation

Feels a little temporal, like a proof of concept. Do you also have higher aims, like to create a flexible coding companion based on language models?