Michael Piotrowski @true_mxp@social.cleverlibre.org

The thing with these large language models is that you don't need insight or scientific discovery, what you need are interns and lots of computational resources.
Look at MT-NLG, the final training ate an estimated 7.8 GWh. That's an energy bill of USD 1,029,922 on a good day.
I guess that's one of the things you can do when you barely pay taxes 🤷‍♂️

"Deutsche Wissenschaftler in Sorge um Twitter"

Beitrag auf Deutschlandfunk zu den Auswirkungen der Musk-Übernahme auf die Wissenschaft. U. a. auch zu ORCID und dem Mastodon-Account von @helmholtz

https://www.deutschlandfunk.de/flucht-vor-musk-deutsche-wissenschaftler-auf-twitter-100.html

Hello! #Introduction

I am an assistant professor in #GameStudies at the Faculty of Arts #LettresUNIL of University of Lausanne #UNIL, Switzerland.

A mathematician interested in #DigitalHumanities and History, my research topics are:
- Video game preservation
- Platform studies
- Video game historiography
- Research creation

With colleagues from HKB, ZHdK and UNIBE, we start in February 2023 a big project on Swiss video games in the 20th Century #SNSF #CHLudens

🖥️ https://yro.ch

So I posted a long rambly thread on Twitter last night sharing my thoughts on the Musk mess. Since people really seem to like it, I thought I'd share it here, too:

So here's my hypothesis on the whole Musk twitter deal. 1/

In these days my thoughts are with all the poor academics who study interaction on the bird site or rely on the bird site for finishing their thesis. What if you just got funding and then you wonder if the site still exists next week? #phdchat #academicmastodon #mastodon

Hallo, wir sind auch #neuhier! 👋🏼

Wir sind das Deutsche Forschungszentrum für Künstliche Intelligenz, kurz DFKI.

Wir werden in Zukunft das #fediverse nutzen, um mit Euch über die gesellschaftlichen Herausforderungen, die sich durch den Einsatz von Methoden der Künstlichen Intelligenz stellen, auszutauschen und zu diskutieren.

CFP: Historical Documents and automatic text recognition - dhCenter

With this special issue of the Journal of Data Mining and Digital Humanities (JDMDH), the goal is to bring together in one single volume several experiments, projects and reflections related to automatic text recognition on Historical documents.

https://dhcenter-unil-epfl.ch/en/2022/11/18/cfp-historical-documents-and-automatic-text-recognition/

#histodons #histodon #histoire #humanitesnumeriques #DigitalHumanites #datamining

#Underline text in different lightweight markup languages:

Emacs #OrgMode , Muse:
_this_

#DokuWiki:
__this__

#Textile, #Jira
+this+

There is no Markdown syntax for underlined text, but #pandoc's #Markdown reader treats the content of spans with class "underline" or "ul" as #underlined:

[important]{.underline}
[nota bene]{.ul}

For those of you interested in how #AI is impacting #education - I have designed a basic story generator, powered by #GPT-3. You, or your students, can add a story title, and/or some opening words, and it will generate a very short story. The more #creative your prompt, the better the result.

It's free to use, with no registration needed, and no adverts (apart from a plug for our book "Story Machines: How Computers Have Become Creative Writers").

https://story-machines.net/

Obviously, #Galactica is still just a large language model, not some sort of universal knowledge engine.

You can't expect it to give correct responses to arbitrary questions about science and the broader world.

Sadly, the way Papers With Code/Meta AI designed the web page may lead to people trusting its language model outputs way too much, as @Michael_J_Black and many others are pointing out.

#NLP #NLProc

The arrogance of thinking that a Large Language Model of all things is going to *organize science* blows my mind.

Call for Sinergia projects
Sinergia promotes interdisciplinary collaboration between 2 to 4 applicants whose project involves breakthrough research. New submission deadline: 17 April 2023! New submission platform: SNSF Portal!
https://sohub.io/u99e

Large documents are easier to handle when split into smaller files. Pandoc will treat multiple input files as if they were one long file.

pandoc -o out.pdf intro.md methods.md

For extra convenience, define the order via filename prefixes, e.g. `01-intro‍.md`, `02-methods‍.md`, and use the shell’s globbing feature:

pandoc -o out.pdf *.md

#pandoc

I've figured out what pisses me off so much about Facebook's Galactica demo.

It's not because people can use to to write bad essays for their homework. There are plenty of large language models that can do that. It's because Facebook is presenting it as something that it most definitely is not.

Facebook is selling it as a knowledge engine, a "new interface to access and manipulate what we know about the universe."

Actually it's just a random bullshit generator.

http://galactica.org