[10:17:12] Forwarded from AaronPaynebot: I am glad i invested [10:17:13] with @Trader_Rice on [10:17:15] binary options, he miraculously [10:17:16] turned my $200 to $2,000 he's the best. Thank you so much for doing a great job, I'm currently enjoying my payout contact him via the link [11:42:54] It would be a bit clunky, but we could make a version of these that accepted the QID as a string. (re @Al: I think neither of these functions would be available yet, but I will check. It is not the functions themselves but their inputs...) [11:46:19] Wikipedias can already call data from Wikidata directly. Can that be passed to a Wikifunctions call? I haven't tried. (re @Al: The way that Wikifunctions works with Wikidata means that some types of data are not yet available. These include numbers and da...) [11:48:09] Yes, but I think Wikidata functions are separately blocked by [[Wikifunctions:Embedded function calls#*Wikidata entity usage for inputs]].* (re @u99of9: It would be a bit clunky, but we could make a version of these that accepted the QID as a string.) [13:41:36] @Dnshitobu thank you for your feedback, I'd like you to put that on https://www.wikifunctions.org/wiki/Wikifunctions:Embedded_function_calls or on the Dagbani Wikipedia Project chat, so that I can link it to the engineers [13:42:03] Thank you very much for trying functions out and giving us feedback, this is what we wanted in the first place [16:45:31] Not too related to Abstract Wikipedia but I wonder if you give an LLM the ability to look things up on Wikidata directly it might hallucinate significantly less [19:46:50] I can imagine an *algorithm* that does it. Probably never perfectly, but better than current LLMs. But an algorithm that is _just_ an LLM probably cannot do it. (re @Feeglgeef: Not too related to Abstract Wikipedia but I wonder if you give an LLM the ability to look things up on Wikidata directly it migh...) [19:47:47] Some LLMs are described as optimized for SPARQL and Wikidata. The most notable one is probably "Spinach". My experience with them was very poor, however. [19:48:56] I was thinking just hooking it up as a "function" like search is done right now (re @amire80: I can imagine an algorithm that does it. Probably never perfectly, but better than current LLMs. But an algorithm that is just a...) [19:52:00] That's one of the biggest problems of current LLMs: all those I've tried look like they are totally not databases, but *only* LLMs, driven not by a specific relationship as in table databases, but only but statistical word completion. That's how it was when ChatGPT made LLMs popular in late 2022, and as far as I can tell, it's still like that now in 2025. I can [19:52:00] imagine that LLMs [19:52:01] will get a better database connection some day, but I haven't seen it happening yet. [19:53:31] That's fair. However you set that up, it will also be ridiculously slow because the amount of Wikidata items you'd want to fetch will take some time (re @amire80: That's one of the biggest problems of current LLMs: all those I've tried look like they are totally not databases, but only LLMs...) [19:56:06] It amuses me a lot that before LLMs came along, computers were known as precise and unambiguous, and people were unexpected and spontaneous, and LLMs kind of flip this. [19:57:57] Language as a whole, unlike math, is inherently imprecise and ambiguous (re @amire80: It amuses me a lot that before LLMs came along, computers were known as precise and unambiguous, and people were unexpected and ...) [19:58:29] Exactly.