That’s what I mean with “if you are sloppy”. If you use an AI as “do this for me but I don’t know what you do” then you have lost and that’s what 99% of people are doing. I use an AI in certain, specific situations only as enhanced pattern recognition engine. This leaves the AI no breathing room to come up with ideas on their own and makes it controllable. In those few situations I do use AI I know every line of code that has been touched and I can ensure every such line does exactly what I want it to. So no exploding, no nail throwing and no chocolate nails since I don’t allow the AI to think, I only allow it to complete a specific pattern receipt I have full control over.
It’s similar as with RegExFu. 99% of people are unable to use RegExFu successfully without turning their code into a mess. I can.
Besides a good software engineer or architect commits only code he has proof-read. Who commits is responsible. That has been the way of trade in all professional projects I had been involved with.
Haha, sure. And who will enforce that? If they use AI well enough, it will go unnoticed.
I see a lot of GSoC applicants are from non-Western background, which is totally fine of course, but it’s also about earning that stipend. That stipend might not be serious money in the US for example, but in some other countries, it is. If a GSoC applicant would use AI to be faster and have another ‘regular’ job as well, that would be an incentive to use AI.
I can tell you that the ban on LLMs has done its job for GSoC students.
Other projects which accept LLM based contributions are drowned in low quality, low effort pull requests and project proposals.
What I’ve seen so far for Haiku seems quite good. And I think we can see the difference between someone who understands what they’re doing, and someone who is just copypasting our questions into a chatbot and copypasting back the answers.
Note also that the ban on LLMs for Haiku is only on LLM generated code, due to licensing concerns since the LLM may regurgitate some existing code that is under a different license, without proper attribution. There are many other ways to use LLMs (such as for code exploration), which are not banned, and would indeed be quite difficult to detect.
It would also be massively stupid. Why risk your GSOC by using AI in Haiku when A) you want to use AI and B) there are other projects that use it
I don’t understand why you think so many GSOC students are faster because of AI. You clearly haven’t reviewed their gerrit submissions…. (They clearly aren’t slop coded.. They actually respond to review comments, like a human would…)