
CS PhD student at @nlpforall, focusing on low-res + morphologically rich NLP. Ulko-🇫🇮-lainen.
Internet
Joined July 2010
- Tweets122
- Following457
- Followers132
- Likes201
Jonne Sälevä retweeted
Check out our paper [arxiv.org/abs/2410.12656] for more details. Huge shoutouts to my amazing collaborators @DCirci, @jonnesaleva, @HaleSirin_, @akoksal_, @bhuwandhingra, @ABosselut, @_dataman_, Lonneke van der Plas.
Jonne Sälevä retweeted
Fixed it.
Jonne Sälevä retweeted
I just feel bad they made Hinton miss his MRI appointment. He's Canadian. His next chance is going to be in like 2 years.
Jonne Sälevä retweeted
BREAKING: The Nobel Prize in Physics has been awarded to ptrblock for “fundamental contributions to physics”
Jonne Sälevä retweeted
Replying to @SeidaSohrabi @helsinkiuni
Suomen valtion virallinen kanta voidaan pukea sanoiksi: free Palestine. Eli kahden valtion politikka. Siinä ei ole antisemitismiä hitustakaan. Palestiinalaiset kuuluvat seemiläisiin kansoihin.
Jonne Sälevä retweeted
Do you like entity names in lots of languages?
Check out @jonnesaleva's paper with me at @LrecColing 2024: ParaNames 1.0: Creating an Entity Name Corpus for 400+ Languages Using Wikidata.
Poster at 11:00 today!
Paper: aclanthology.org/2024.lrec-m…
Data: github.com/bltlab/paranames
Jonne Sälevä retweeted
We are pleased to announce that CS faculty member Duygu Ataman (@_dataman_ ) has received a research award from Microsoft's Accelerate Foundation Models Research (@AFMResearch) program. Congratulations! microsoft.com/en-us/research…
Jonne Sälevä retweeted
10 years ago, WFST-based methods were the norm for speech processing (think, Kaldi).
Since then, end-to-end models have become quite the rage --- they are simple, do not require much domain expertise, and you can train a PyTorch model for a new task over a weekend. ⚡️
1/n
Jonne Sälevä retweeted
I think one of my weird takeaways from looking at LLMs is that language is more strategic than was generally assumed and extremely rich in implicit information.
Jonne Sälevä retweeted
Depression and your PhD Thread
Getting real depression while working on your phd is a really easy ~40% probability, I got it too, and here is my experience.
Disclaimer: I am not a medical professional, so consult with your doctor.
Jonne Sälevä retweeted
I looked a little into the Gzip OOD results and there seems to be another big problem: train-test overlap.
E.g. DengueFilipino has the same train and test set. KirundiNews has 90% overlap...
Still nice to see people revisit old ideas and the use of information theory for ML :)
Jonne Sälevä retweeted
PhD opportunity on keeping Indigenous languages strong in northern Australia, through storytelling, art, landcare, social exchange, tourism, ... under Indigenous leadership. $40k/year domestic scholarship. Please send expressions of interest to me at [email protected]
Jonne Sälevä retweeted
President Niinistö, I remember a phone call you and I had ahead of your visit regarding your joining NATO.
It took me three seconds to offer my strong support.
Today, following the fastest ratification period in modern history, I'm proud to say Finland will make NATO stronger.
Jonne Sälevä retweeted
Meanwhile on Earth 2, David Blei is testifying before Congress about the existential risks posed by topic models, after LDA-4 takes the world by storm.
Jonne Sälevä retweeted
"you are not spartacus. you are just another person. a person without substantial wealth or powerful friends"... said the "tenured professor at an R1 university".
Jonne Sälevä retweeted
We are visualizing our data all wrong.
Typically, academic graphs show statistical confidence intervals or standard errors. This paper shows that everyone from doctors to tenured faculty misread these graphs, assuming effects are larger. Show variability! jakehofman.com/pdfs/illusion…
Jonne Sälevä retweeted
Prof. Gil Strang's last lecture is ... Monday! Watch it via live stream: grinfeld.org/strang/
Jonne Sälevä retweeted
People are testing large language models (LLMs) on their "cognitive" abilities - theory of mind, causality, syllogistic reasoning, etc. Many (most?) of these evaluations are deeply flawed. To evaluate LLMs effectively, we need some principles from experimental psychology.🧵
Jonne Sälevä retweeted
Finnish president @niinisto joining soldiers for the Kontio 22 military exercise.
He spent the night in one of the tents as the rest of the soldiers.
🇫🇮

































