I reply to Jason and Jeremy below.


On Fri, May 9, 2025 at 4:17 PM Jason McKesson wrote:
>
> If you want to do that, that's fine. But we don't need updates, as
> this is outside of the scope of the ML. Feel free to advertise your
> results of your work, but that's not really what this place is for.

On Fri, May 9, 2025 at 4:50 PM Jeremy Rifkin wrote:
>
> There's nothing to discuss. If you think that would be useful, make it.
>
> As others have pointed out, google is more than sufficient most of the time,
> especially since these are technical subjects which typically have easily
> searchable technical terms. One trick you can use is to add
> "site:https://www.open-std.org/jtc1/sc22/wg21/docs/papers/" to your google
> search. If you want an AI tool for searching, you're free to make such a thing
> but evidently the interest from others on this mailing list is minimal.


I get it, the two of you don't want a running commentary of the
progress I make on putting together an offline program that uses AI to
do RAG(Retrieval augmented generation) on the six thousand papers.

If anyone wants to try this themselves, I've downloaded the six
thousand papers to my own webspace and compressed them, so you can
download them from these links:

    625 megabytes --
http://virjacode.com/download/all_cxx_papers_2025_05_10.tar.xz
    834 megabytes -- http://virjacode.com/download/all_cxx_papers_2025_05_10.zip

Here's some suggested YouTube videos for running an AI offline and
making it do RAG:

    https://www.youtube.com/watch?v=gigip1Pxf88
    https://www.youtube.com/watch?v=AHlx91CKmmc
    https://www.youtube.com/watch?v=WxYC9-hBM_g

I've decided to fork the "llama.cpp" repo up on Github, as it produces
a shared library "libllama.so / llama.dll / llama.dylib" which you can
use to interact with an LLM offline such as DeepSeek. Here's my new
Github repo for this:

   https://github.com/healytpk/ai-cxxpapers/

I'm going to add code to this repo for a GUI application that works in
two steps. The first step is that you ask a boolean question (i.e. a
Yes-or-No question) in order to whittle down the list of papers. An
example boolean question would be:

    "Does this paper suggest adding a new cv-qualifier to the language?"

You click "Next", and after a minute or two, a list control will be
populated on screen with the paper numbers that answered yes, so
something like:

    P1234R5
    P1783R2
    P2748R9
    P2923R2
    P3105R1
    P3122R4

Then in the second text box, you can ask a more specific question (not
necessarily a boolean question) such as:

    "Make a list of all the cv-qualifiers that this paper suggests
adding to the language."

And so then a multiline textbox on screen will come back with something like:

    P1234R5 -- restrict
    P1783R2 -- constarr
    P2748R9 -- noexcept
    P2923R2 -- var
    P3105R1 -- code_mem
    P3122R4 -- noexcept

I'm very certain that I myself will find this new program very useful.
Anyway I won't give a running commentary . . . I'll just post the
final link to download it when I have it done.