What do you think about DeepWiki for ParaView?

ParaView Support
,
1

Hi,

Recently, my web searches frequently recommend DeepWiki for programming topics. I noticed that ParaView is indexed in the wiki. At the bottom of the page, you can type a question. There are two main sections in the generated answer: Fast and below it Codemap. The Fast section is more an overview, with some extracted code, while Codemap writes concrete code for your question. I gave it a try by asking how to write a custom application? Here is the result. I chose this question because I have some experience in this topic so that I can partially validate what the AI generates. My impressions:

  • I am quite satisfied with the Fast section. It successfully recognized that the paraview_client_add CMake macro is responsible for building the client executable, and classified the parameters as required and optional. It also recognized that the ParaView-specific pqApplicationCore must be provided in addition to the standard Qt class QApplication. It “understood” that the ParaView GUI can be thought of as a pre-made client and showed it as an example. Finally, I appreciated that all the claims it made are supported by actual code of the ParaView code base. So the Fast section can be useful to discover unknown topics.
  • The code in the Codemap section is on the other hand not very usable in my opinion. It is overly verbose. One can write much more succinct code by following the examples of Examples/CustomApplications (which the AI failed to find).

My questions to ParaView core developers or people with deep knowledge of ParaView:

  1. Do you think that the main page of the wiki is technically correct (I am not asking to read everything, just a few sections)? If not correct, at least it properly grasps the main concepts?
  2. If you ask questions in the prompt, are you satisfied with the results?
  3. Do you think that DeepWiki is useful for ParaView programmers, or it is just the usual AI slop?
2

Looks like slop to me, some info is correct, some is not, global architecture is wild, BSD license is not respected at all and they point github repo, not gitlab repo.

Don’t rely on such system, because it means proper doc will be left to rot imo.

3

Thanks for the validation. This is my opinion too, in general for LLM answers.

they point github repo, not gitlab repo

Probably because somebody fed the GitHub mirror to DeepWiki. GitHub is simply more popular than GitLab, as demonstrated by the almost ten times more stars (not that it measures the value of a software).

BSD license is not respected

What do you mean?

because it means proper doc will be left to rot imo

I.e. people will use the AI-generated nonsense instead of reading (and potentially contributing to) the official documentation?

4

Developers are writing the docs (that are actually used to feed such llm system) however, if nobody reads them, why bother writing them, right ?

And indeed, user reading docs find missing info, inconsistencies and actually fix them, which is good for everyone.

5

As usual with AI tools, data is scraped and then provided to user without providing the source of the data nor respecting the license of the data, which here is BSD licensed which would mean mentioning the Kitware copyright everywhere.

1 Like