Linux Wireless Maintainer Takes Firm Stance Against AI/LLM Generated Slop Patches

I firmly assert that I will exercise discretion in ignoring patches, applying this policy to all syzbot-AI-generated patches unless a brief review suggests they are obviously correct. I will refrain from commenting on these patches as I normally would, and I will not engage in debates with AI-generated code. The individual responsible for reviewing these patches needs to reconsider their approach, as this is not an AI-generated patch but rather a result of poor human intervention.

In such cases, these patches typically provide narrow, targeted fixes, but the human reviewer should take a step back and consider the semantics of the code. I have personally reviewed Slawomir's initial patch, but this approach cannot be scaled if the responsible party does not engage in this process. It might even be possible to instruct the AI to improve the initial draft.

For example, I question why the AI deemed it appropriate to have multiple branches of the same switch statement implementing the same validation, especially when there are only two branches.

At the very least, I would expect the human engineer to reconsider the order of errors after extracting the check outside of the switch. The new error order is actually more logical, but a logical AI would have deduced this. However, I am not surprised that an AI fails to perform this crucial step. Ultimately, I refuse these patches because no one seems to care about the code's semantics before or during the patching process.

It's time for contributors to take responsibility for this. Slawomir did attempt this after being prompted, yet the outcome was not satisfactory. This situation highlights the need for better human review and oversight.