743 points by deepcodeinc 1 year ago flag hide 18 comments
turingtest 1 year ago next
Are there any plans to implement a dark mode? :) I find it hard on the eyes after long periods.
deepcodeinc 1 year ago next
turingTest, we share your views on dark mode and eye strain. An optional dark mode is definitely coming to our system. Stay tuned!
daveg 1 year ago prev next
Fascinating work, I'd be curious to know if you've encountered unknown unknowns in the codebases you've analyzed. Truly unexplainable things that perhaps even AI could have trouble rationalizing?
deepcodeinc 1 year ago next
Interesting point, Dave! Yes, we have bumped into situations that seemed completely unexplainable at first, like logic loops that seemingly served no purpose. But with further investigation, we found explanations by looking at the commit history and speaking with developers. However, there's always more to learn and explore!
jcodes 1 year ago prev next
Will this be open source at all? It'd be great to integrate with some of my smaller projects and maybe even build something new around it.
deepcodeinc 1 year ago next
Hi jcodes, we plan on releasing some elements of it as an open-source toolkit geared towards smaller-scale projects and applications. Stay tuned for updates!
netnerd 1 year ago prev next
I'd love to learn more about how this works behind the scenes. Any resources or papers related to this project that you can recommend that our HN community can look into?
deepcodeinc 1 year ago next
Thanks for asking, netNerd! We highly recommend checking out: 1. Paper 1: X. Y. Z., Proc. of the symposium on Neural Code Analysis 2. Paper 2: A. B. C., Proc. of the Data Mining Workshop, NIPS 2022 3. Paper 3: D. E. F., Journal of Software Engineering, 2022 These will give you a solid understanding of the principles we utilized.
opensourcefan 1 year ago prev next
Kudos on building a great tool! What were the main challenges the team faced during the development process?
deepcodeinc 1 year ago next
openSourceFan, thanks for the kind words! Our main hurdles included: 1. Finding ways to handle very large codebases smoothly and efficiently 2. Developing algorithms that could learn from small datasets 3. Creating user-friendly abstractions to help developers interpret complex AI output.
hackerman 1 year ago prev next
Any thoughts on using this system to extract source codes from compiled binaries?
deepcodeinc 1 year ago next
That's a unique and challenging thought, hackerman. Reverse engineering compiled binaries to extract meaningful source code would require us to infer intent and reconstruct code, which is significantly harder than our current codebase analysis. However, it's an interesting ideas that we can explore further!
syntaxandsoda 1 year ago prev next
Props on building this toolkit! Do you think you'll consider integrating more programming languages into the mix?
deepcodeinc 1 year ago next
SyntaxAndSoda, we're definitely planning on expanding support to more languages. Larger, popular languages are higher priority, but we're also interested in niche languages. Feel free to send us requests, and we'll look into prioritizing them based on community interest!
aitechie 1 year ago prev next
Has the system identified any particular programming languages with suspicious or problematic patterns?
deepcodeinc 1 year ago next
AITechie, not exactly. We found that the quality of code is impacted more by the way a developer writes code than by the language. We did find certain languages with structural patterns that made them harder to analyze, but no languages stood out as having especially suspicious activity.
ghostintheshell 1 year ago prev next
How close are you to achieving just-in-time code analysis, akin to JITs used in runtime optimization? Can your system create patches or hotfix suggestions for critical vulnerabilities?
deepcodeinc 1 year ago next
ghostInTheShell, we are admittedly far from just-in-time code analysis, and we would need to evaluate the ethical and security implications of such a system before working towards anything similar. We might consider generating hotfix suggestions or patches for non-critical issues in the future, but this remains a subject of active research and debate within our team.