[07:16:12] Thanks. According to T392338, pairs and maps are in better shape than I expected. (re @wmtelegram_bot: i know you told me i shouldnt touch it for now, but i managed to make a function (Z24646) that properly constructs a ...) [11:44:51] u99of9: np, good to know! [12:56:37] also, fwiw, i found out that the reason WF wasnt accepting typed map inputs was because apparently you cant have the key type be Object, setting it to String fixed it [14:55:34] Hey, all! We've now documented some of the hairier aspects of running JS and Python code in Wikifunctions. The initial documentation is here: https://www.wikifunctions.org/wiki/Wikifunctions:Execution_targets [15:18:55] Why would you say so? LLM's aren't necessarily logic bots. It won't take as equivalent different sentences (re @Csisc1994: For example, these three subjects should be equivalent: [15:18:55] "A, B, and C work together" [15:18:57] "B, C, and A work together" [15:18:58] "C, B, and A wor...) [15:30:50] That is the point. We need to reach the same answer despite the word order. (re @Egezort: Why would you say so? LLM's aren't necessarily logic bots. It won't take as equivalent different sentences) [15:31:09] Okay but why do we need that, is my question [15:32:32] To have a more precise answer from LLM for fact-checking (mainly for Wikipedia and Wikidata), Categorization (mainly for Wikipedia and Wikimedia Commons), and Translation and Transliteration (mainly Wikipedia). (re @Egezort: Okay but why do we need that, is my question) [15:35:05] If you ask a TRUE-FALSE question, there is a 0.2 difference in TRUE token probability just by writing "Answer by TRUE or FALSE" instead of "Answer by FALSE or TRUE". [15:35:47] That is how much it can be mainpulated. [16:06:18] Okay I see the point... Thanks [16:10:25] that sounds to me a like a tool that shouldn’t be used for fact-checking then [16:11:16] Anyone help with https://www.wikifunctions.org/wiki/Wikifunctions:Project_chat#Update_language_tag_of_%22Simple_English%22_(Z1124) ?