Michael Piotrowski @true_mxp@qoto.org

This is a big challenge with prompt leak attacks generally: the model just guesses what word should come next, so once it starts spitting out pieces of its own prompt it's perfectly capable of inventing new prompt segments out of thin air

And anything it invents will look convincing, because the whole point of large language models is to generate stuff that looks convincing!

On February 15 at 18:15 I will talk about "Sentiment Analysis for Latin: lexicons, annotation and automatic approaches" within the #DigitalHumanities colloquium at Freie Universität Berlin. No registration needed, just connect here: https://meet.in-berlin.de/dh_colloquium_fu_berlin

Scientists will actually tell you what they don’t know.

Some serious research in ChatGPT capacities and limitations https://arxiv.org/abs/2302.04023?fbclid=IwAR1Q3v7jty7nlejw1d4KEW4A4vxyPNKyZ3OhkUnTzM04vFOZ1fxAo8B6FcA

Berghotel Schatzalp near Davos was built between 1898 and 1900 as a sanatorium and was one of the inspirations for Thomas Mann's Novel "Der Zauberberg". It was one of the first reinforced concrete buildings in Graubünden and featured many modern technologies such as floor heating, an elevator and a mail and telegraph room. Since 1954, it is a hotel with 92 rooms are spread across three floors. #architecture #photography #Switzerland

#whataweekhuh

Vor 60 Jahren starb Carl Stucki – oder wie man vom #Idiotikon-Redaktor zum Überwacher des Waffenstillstands nach dem #Koreakrieg werden kann: https://de.wikipedia.org/wiki/Carl_Stucki

I wrote a longer-form piece on post.news about the concept of operationalization in AI and the way it disenfranchises communities from making decisions that the systems that govern their lives.

If you want to read it there: https://post.news/article/2LR4k0jt685pzd6uiGeHJlnBwo2

I'll also serialize it, below.

Introducing tootfinder.ch

Proof of concept of an opt-in global Mastodon search. Tootfinder indexes recent toots of all user that want to get indexed. If you are not interested, just do not join the index.The feeds are indexed in a SQLite database and deleted after 7 days.

Looks like our #DigitalHumanities web app #ezlinavis (https://ezlinavis.dracor.org/), originally intended for extracting co-occurrence network data from literary texts, is also used in medical research. 😯 @cmil

https://doi.org/10.1007/s00432-022-04200-0
https://doi.org/10.21873/invivo.12976

RECEPTIO's version of #receptiogate:

"In December 2022, we were the victim of a vicious defamation campaign via the web, which shut us down for a month. After a legal investigation that established that all the accusations against us were unfounded and the result of the ravings of a sick mind (to which haters gave credence), in February 2023 the centre resumed its cultural activities"
[https://www.receptio.eu/story]

The lack of a full stop after the last sentence suggests that the story isn't quite finished yet...

Im neuen SAGW-Newsletter gelesen: «Universität Zürich schafft unbefristete Stellen im Mittelbau», sog. Lecturer-Positionen in Lehre und Forschung. Die Stellen sind für Postdocs vorgesehen, Verbesserungen gibt es aber auch für Doktorierende und Assistierende. Bleibt nur die Frage der Finanzierung:
https://www.sagw.ch/sagw/aktuell/news/details/news/next-generation-uzh

Les différents modes d’écriture de l’histoire et leurs effets sur les pratiques historienne. https://www.infoclio.ch/de/les-diff%C3%A9rents-modes-d%E2%80%99%C3%A9criture-de-l%E2%80%99histoire-et-leurs-effets-sur-les-pratiques-historienne?utm_source=dlvr.it&utm_medium=mastodon #infoclioevent

We've made it easier to use the #Lua interpreter in #pandoc 3: calling `pandoc lua` (or invoking pandoc via a symlink named `pandoc-lua`) lets pandoc behave like the default `lua` binary, but with all of pandoc's libraries loaded. E.g.,

echo 'print(PANDOC_VERSION)' | pandoc lua

prints the version, and

pandoc lua script.lua

runs the given script.
#LuaLang

https://pandoc.org/MANUAL#running-pandoc-as-a-lua-interpreter