- research, science, technology, music, personal opinions, etc.
Are IT systems beyond LLM wikis necessary? Could an organization run almost entirely on a large language model, a versioned wiki, authentication, and a few deterministic services? Internal emails, invoices, administrative documents, intranet pages, slides, notes, signups, schedules, timesheets, vacation rules — much of organizational life consists of documents moving between authorized people.…
I have previously, in April 2023 and in October 2023, asked ChatGPT about the Danish sentence Børnehavebørn leger med sandkassesand , where there are peculiar repeated compounds. At that time, it had difficulty in recognizing the repeated pattern. Now it recognizes unprompted for that particular aspect that there is a comical repetition. But it [ ]
The Scholia web app at https://scholia.toolforge.org/ has recently experienced an unusual load. The above image is a screenshot from the Wikimedia Grafana for Scholia available as https://grafana.wmcloud.org/d/TJuKfnt4z/tool-dashboard?orgId=1 var-namespace=tool-scholia var-cluster=P8433460076D33992 from=2025-10-27T05:25:59.007Z to=2025-10-27T21:47:59.006Z timezone=utc viewPanel=panel-27. At times…
Generative artificial intelligence with strong large language models (LLM) as available in ChatGPT, Gemini, Copilot, etc. or in locally running LLMs poses a challenge for the evaluation of student performance and teachers move to oral or pen-and-paper exams to counter student use of the LLM tools. In universities, where classes may be large, say 150, [ ]
My previous post asked Was there an AI winter between 2020 and 2023? and was based on first occurrence of Danish AI -compounds in Wikidata. The SPARQL results seem to indicate that there were no new neologisms between (approximately) 2020 and 2023. I have added some more AI -compounds to Wikidata as well as pointers to attestors with [ ]
Wikidata lexemes can record compounding and attested by , and this I have done for some of the Danish words where AI is a part. For some reason the Danish language uses the English abbreviation for artificial intelligence instead of abbreviating the Danish kunstig intelligens . So far I have found 30 Danish words prefixed with AI- . [ ]
Wikidata now has a good number of compounds (including affixiations) specified for lexemes in its lexicographic namespace. Over 13,000 Danish lexemes are specified to be compounds. Over 2,300 Danish lexemes are specified to be suffix-derived lexemes. Different analyses can be made of this graph, e.g., with the Python library NetworkX. Graph analysis finds that interfixes [ ]
Wikidata Query Service (WDQS), the triple store (Semantic Web database) with a SPARQL endpoint (the Web service for querying) for Wikidata, is scheduled to be split into two parts i 2025: One for the scientific articles et al. and one for the rest. Our Web application for Wikidata scholarly profiles, Scholia, relies on WDQS for [ ]
As noted on the Web, large language models (LLMs) have a problem with the question: How many r are there in the word Strawberry ? . Often LLMs fail and report that there are 2 r in the word, while there are actually 3 r s. This is probably due to the tokenization in the first part of [ ]
ChatGPT and other large language model chatbots may be good at answering questions, e.g., back in December 2022 I tried the new ChatGPT on the Danish national citizenship test (indfødsretsprøven) with good results. It mostly failed on the questions which pertained to new information after the training deadline. But how good is ChatGPT to answer [ ]