Title
#share-your-work
Kiril Videlov

Kiril Videlov

11/25/2022, 3:07 PM
I saw the ‘AI found a bug in my code’ on HN and I wanted to try this out on my code. Since the tool was not open sourced, I decided to build it from scratch myself 👨‍💻😅. So I present to you Suspicious (https://github.com/sturdy-dev/suspicious) It’s a fully local CLI app which analyses your source code file. It then shows you specific parts of the code where it would have done something else. Demos & screenshots in the GitHub readme. Let me know if it catches anything good for you 🙂
Duncan Cragg

Duncan Cragg

11/25/2022, 3:53 PM
I tried it - there's a lot of downloading! And to be honest, it didn't highlight anything that worried me. Sorry to seem negative, but there it is! 🤗
3:53 PM
What is it downloading after running
sus
?
Kartik Agaram

Kartik Agaram

11/25/2022, 5:10 PM
Duncan Cragg

Duncan Cragg

11/25/2022, 11:32 PM
It was 500Mb
11:33 PM
sus
uses the decoder of UniXcoder, specifically the unixcoder-base-nine checkpoint. What's cool is that it's only 500 MB and ~120M parameters, which means it's quick to download and fast enough to run locally.
11:34 PM
Ah ↑↑
b

Ben Byford

11/27/2022, 1:21 PM
Super interesting. I would love a Autopilot alternative
Kiril Videlov

Kiril Videlov

11/28/2022, 12:21 PM
Hey, sorry, I will highlight more prominently in the readme that a model is being downloaded. It’s kind of the tradeoff for being able to execute everything locally.
12:23 PM
Building this application got me to realise that the way current models (this and other ones like Codex used by Copilot) can mostly serve as typing automation because of the way tokenisation is done
b

Ben Byford

11/28/2022, 12:40 PM
As apposed to something else? A static recommendation on all the code base?
12:42 PM
Would be nice to get recommended differing ways of doing the same thing…