You're reading the public-facing archive of the Category Theory Zulip server.
To join the server you need an invite. Anybody can get an invite by contacting Matteo Capucci at name dot surname at gmail dot com.
For all things related to this archive refer to the same person.
Jan Pax (Dec 31 2025 at 11:06):In the book introduction to the theory of cooperative games by Peleg and Sudholter I'm looking for the paragraph which roughly says:
the reduced game is a hypothetical game.
AI from claude claims it is on the page 112, but searching in pdf doesn't show any occurrence at all in the entire document.
David Michael Roberts (Dec 31 2025 at 12:09):Why are you trusting an LLM to correctly generate a citation, let alone to a specific page?
Jan Pax (Dec 31 2025 at 13:41):I remember having read this collocation in the book but I cannot locate it now.
Jan Pax (Dec 31 2025 at 13:42):I have provided Claude with the pdf of the book but it answers incorrectly.
Ryan Wisnesky (Dec 31 2025 at 18:52):Even if you give an LLM data and a correct prompt, there is zero guarantee it will produce correct results, by definition of how LLMs work. (A search for a string by an LLM in a document may find it, may miss it, or may find something else entirely.) Everything that an LLM generates must be checked. But a lot of people seem to be siding with LLMs over librarians when it comes to these made-up citations: https://gizmodo.com/librarians-arent-hiding-secret-books-from-you-that-only-ai-knows-about-2000698176
James Deikun (Dec 31 2025 at 20:43):There's a certain amount of inherent unreliability, true; just like there would be with a human if you told them to read a whole book and tell you what page a certain sentence was on, and they weren't allowed to answer until they finished the book (I think this is how this task would end up being set up on existing chatbots).
But the biggest problem with using LLMs for research is how obsequious they are. They really go out of their way to figure out what you believe and confirm it, even if they have to make up citations that don't exist or say things they're perfectly capable of realizing aren't true. You can make up for this for some extent with prompting, especially if you have enough access to override part or all of the system prompt, but it's also part of post-training so it's hard to erase entirely with a prompt. It also helps to never be even slightly assertive in stating something you are not absolutely certain of, but even leading questions can cause this behavior so it's not so easy to avoid.
Jan Pax (Dec 31 2025 at 20:46):But I remember personally to have read about a "hypothetical game" in that book but now I cannot locate it.Any chance for me to re-locate it ?
James Deikun (Dec 31 2025 at 21:08):Unlike many PDFs from your source, this one does have actual text; there are no occurrences of "hypothet" and none of the occurrences of "ical" are from hyphenated occurrences of "hypothetical". There are no entries for hypothetical games in the index or TOC. Thus if it really was from this book, "hypothetical" must be meant in the ordinary sense. On page 22, describing the "reduced game" for the first time, we have:
Thus, is not a game in the ordinary sense; it serves only to determine the distribution of to the members of .
This may have been phrased differently in a different edition, using the word "hypothetical" to describe the reduced game. Alternatively, you are misremembering which book you saw this in.
David Michael Roberts (Dec 31 2025 at 23:14):I would recommend just searching google scholar or google books for "hypothetical game" cooperative games to use an actual search engine that has indexed scholarly literature. I get no obvious hits, so you may well be misremembering