[manojpv] Reviewing exercise submission

[bossbeagle1509]

Reviewing exercise submission:

• Gitlab username: manojpv
• Repository: https://github.com/bossbeagle1509/sse-responsible-ai-assignment
• Pull Request: I acknowledge the use of RAI (https://rai.uni-stuttgart.de/) to imple… bossbeagle1509/sse-responsible-ai-assignment#1

Generating with RAI

• Pro: It seems to have gotten a fair part of the heavy math right.
• Con: The biggest downside is the lack of input validation and error handling—things like checking if nx >= 2 (to avoid division by zero), alpha > 0, or handling missing config files gracefully. This makes it brittle for real-world use, where bad inputs could crash it unexpectedly.

Code Review

• As a Computer Science student, I focused on the code itself and system robustness.
• Copilot focused on mathematical and physical errors or improvements like an expert on simulation software domain.

What I learned

AI can be good at project scaffolding and doing some of the heavy lifting, but in the end someone does need to validate and review the output.

Examples from my challenge project

• Add general lazy evaluation for all blosc2 functions Blosc/python-blosc2#496
• Add saving lazyudf Blosc/python-blosc2#532

Maintainers seem to be focussed on maintainability and correctness of their code as well as not attempting to do too much per merge.

[MakisH]

Hi there, sorry for the delayed feedback. This submission has several weaknesses, but is still valid. Some comments:

• There is an AI disclaimer in the commit message, including the prompt.
  • However, I would have used a descriptive commit message, PR title, and a feature-oriented description, not just the disclaimer.
• There was a single prompt, which seems to have provided enough details. One could also build that step-by-step and add more details.
• The human review was rather short, lacking inline suggestions (comments in a copy of the code are not the right approach). See the GitHub documentation for some examples, or how Copilot commented on the same PR. The summary was generally fine, but I would have started with some more personal opening.
• There is an AI review with Copilot, but there was no follow-up. You could comment on the suggestions and apply or close them. You could even ask Copilot to implement the feedback for you.
• The comparison between the human and AI reviews is fine, rather short, but most importantly 1-to-1 copied from [shayo] Reviewing exercise submission #16 (comment) without any reference. This is wrong, but in this case only affects a very small part of the submission.
• The examples from python-blosc2 have review discussions, but the comments are rather vague (generally true for every project).

You have now tried different tools and looked at how others in your challenge project review PRs. With practice, everything will become easier and clearer.