One can solve it as RAG (Retrieval Augmented Finetuning), where the LLM uses natural language processing to help formulate and invoke the search queries and filter the found entries. With a defined API interface to directly access the proposal archive. or with Finetuning, where the LLM learns all discussions and proposals, and has the knowledge as if it was trained with it. Often some layers of the LLM would change weights, while others/most stay the same. or with Creating summaries of all C++ ideas discussions and proposals and put them into the context window of the LLM.   As all proposals/discussions are available on the website, each of those should be simple to do.   You could also consider to add other materials, like the C++ draft (if the license allows) or the Core Guidelines or some language related Blogs (if their license allow). Make it name the sources/references and recheck arguments it makes against sources by itself to limit hallucinations.   I believe one can make it useful enough that some proposal authors would even consider the feedback from the model in addition to human feedback to get new references, new ideas or some possible weak points for improvement.   -----Ursprüngliche Nachricht----- Von:Thiago Macieira via Std-Proposals <std-proposals_at_[hidden]> Gesendet:Do 08.05.2025 17:02 Betreff:Re: [std-proposals] Dedicated website with AI that has processed all papers An:std-proposals_at_[hidden]; CC:Thiago Macieira <thiago_at_[hidden]>; On Thursday, 8 May 2025 16:43:24 Central European Summer Time Frederick Virchanza Gotham via Std-Proposals wrote: > AI is fine for some uses though, like perusing a catalogue of three > thousand papers written about proposals to change to C++ programming > language. Basically I just want the AI to 'crawl' the papers so that I > can ask simple questions like, "Has anyone ever written a paper to > allow the calling of non-const methods on a const object if we're > dealing with an Rvalue?". > > The AI would then direct me to papers that I myself can take a look at it. If you know the keywords to search, just use a search engine. You're quite likely to know keywords because, as a technical problem, there are specific terms for specific things. https://www.google.com/search?q=site%3Aopen-std.org+non-const+rvalue Quite a few search engines are now integrating natural language processing to find things for you. Maybe try that. If you want to train an LLM on the papers (which are all available) and host it, go ahead and do it. -- Thiago Macieira - thiago (AT) macieira.info - thiago (AT) kde.org  Principal Engineer - Intel DCAI Platform & System Engineering -- Std-Proposals mailing list Std-Proposals_at_[hidden] https://lists.isocpp.org/mailman/listinfo.cgi/std-proposals