Show newer

My colleague Julia Ecker who is not on the Fediverse spent a lot of time creating a new promo video for our chemistry department. It's a nice journey through the whole breadth of research and labs we have here at ETH Zurich. Check it out at youtube.com/watch?v=1vCVhRNQmm

@mathowie I had downloaded it and used it successfully. Then, when I was somewhere in the wilderness with no internet and wanted to analyze some bird sounds, it wanted me to log in again. Shitty UX design.

Often when I post something about my research on software teams somebody asks in a hostile kinda way if I even code (I do but I don't answer that question from strangers directly on principle). Now that I'm writing a paper on management I look forward to seeing if people ask if I even manage.

@MediaLawProf So are fake eyelashes and fingernails banned, too?

Here is the belated picture of the complete meal: Pork loin roast in a red wine sauce seasoned with ginger, coriander seeds, cloves, curry powder and cinnamon; 100 curry; carrots with cardamom; semolina polenta.

Show thread

First time I've seen this:
"[publisher] will donate 5% of your [APC] fee to the charity of your choice. You will be asked for your charity preference on the order form you complete."

I dislike this. If the cost of running the journal requires an APC, fine. But why should you build in a 5% donation to my favorite charity? Plus there's the extra administrative costs to do this. How do funders feel about their grant funds being spent on APCs with charitable kickbacks?
#scholcomm

Hey out there. I have a dumb PubMed question on term explosion with subheadings: In the 4 queries on the first pic (see ALT-Text for a copyable version) the numbers for the exploded codeine search are much higher than the combined numbers of the other three.
The behavior is different if I don't work with subheadings (cf. second pic). Then the numbers of the non-exploded main term and two more specific terms add up to a little more than the exploded main term. This makes sense as I would expect some papers to be on both levels of the hierarchy.
Anybody out there got an explanation for the behavior in the first picture.

In case anybody was wondering what 229 cloves of garlic look like. Here you go.

That feeling when you're in the restaurant with your small kids, a glass shatters at the next table and the waitress heading straight for you shows a really confused expression because she cannot see any shards on your table.

Just read a great idea that we should start naming mass shootings after the NRA-owned lawmakers whose districts are closest to the site of the slaughter that their negligence condoned.

So with that in mind, for the recent Louisville Bank Shooting, I hereby nominate Rep. Thomas Massie.

His district borders Louisville and he loves guns so much that he dedicated his Christmas card to them.

From this day forward the Louisville Bank Shooting will be known as THE THOMAS MASSIE MASS SHOOTING!

There it is: Germany's second big public broadcaster is teasing, that it will start posting on #Mastodon tomorrow as @ZDF That account has 3200 follower, on #Twitter its 1,4 million.

I'm sure it will start to grow quickly and get a headstart on the other big public broadcaster #ARD, which has labeled its first account a test (@NDR).

Of course it would also be a headstart on the big public broadcasters, that are arguing with #Twitter.

#NPR #PBS #TwitterTakeover #ElonMusk #SocialMedia #BBC

There's also a rapidly-expanding set of GPT variants trained on specialized data sets, like this one trained on PubMed. This definitely is more than an afternoon of work or whatever, but if you want it to know Stata, training it on Statalist is realistic. github.com/stanford-crfm/BioMe

For moderately sized stuff like if you want it to speak in the language of a particular technical book, you can feed it that book as an embedding via LangChain; this I've done and it's not suuuper hard to figure out (although pls don't ask me for help). github.com/hwchase17/langchain

Show thread

I have not yet pursued this myself, but if you're trying to use GPT for [niche topic] and it doesn't work that well because it's not a popular enough topic (like Stata, or the lingo of formal causal modeling), it doesn't appear to be THAT difficult to train your own.

@ct_bergstrom I screenshotted this toot a while ago as it resonated with me. Regarding the part in quotation marks, I assume you are speaking and not Bruno Latour, right? I am preparing a session on text-based lit search and I would like to get the attribution right.😜

Show older
Qoto Mastodon

QOTO: Question Others to Teach Ourselves
An inclusive, Academic Freedom, instance
All cultures welcome.
Hate speech and harassment strictly forbidden.