固定されたツイート
LingVis.io
66 posts
LingVis.io がリツイート
Want to explore #wordembedding properties? Here are some ideas for data preparation steps✂️, interesting additional features🍩, and some relevant references✍️: embedding-framework.lingvis.io With
@melassady 😊 #nlviz #ieeevis #LLMs
English
LingVis.io がリツイート
Our visualizations show that larger models, e.g., ChatGPT and Bard, are less biased than their smaller counterparts. Are the models getting better or the engineering steps used to adapt the outputs? 🤔@melassady #ieeevis #vds 👉prompt-comparison.lingvis.io
John Alexis Guerra Gómez @johnguerra.bsky.social@duto_guerra
Very interesting work: Visual Comparison of Text Sequences Generated by Large Language Models R. Sevastjanova, S. Vogelbacher, A. Spitz, D. A. Keim, M. El-Assady #VDS Symposium @ieeevis Demo and paper 👉🏼prompt-comparison.lingvis.io
English
LingVis.io がリツイート
What biases are encoded in texts generated by #LLMs? Our workspace helps to explore stereotypes encoded in prompt outputs and detect unexpected word associations. @melassady will present our work
@VisualDataSci🥳 Paper and demo: prompt-comparison.lingvis.io #ieeevis2023 @ieeevis
English
LingVis.io がリツイート
How should we compare the embedding spaces of large language models? Did language model adapters (@AdapterHub) introduce bias? Join us in the current #ieeevis session on comparison in Oklahoma Station 1. #vis22
Rita Sevastjanova@RSevastjanova
Our @ieeevis paper shows how to visually compare embedding spaces produced by adapted language models. Thanks, @AdapterHub, for the variety of adapters for evaluation!🥳And a big thank you to my co-authors E.Cakmak @ravfogel @ryandcotterell @melassady!😀 adapters.demo.lingvis.io
English
LingVis.io がリツイート
Our @ieeevis paper shows how to visually compare embedding spaces produced by adapted language models. Thanks, @AdapterHub, for the variety of adapters for evaluation!🥳And a big thank you to my co-authors E.Cakmak @ravfogel @ryandcotterell @melassady!😀 adapters.demo.lingvis.io
English
LingVis.io がリツイート
If you are interested in #languagemodels, you might want to check out this interactive article on gender bias in BERT 😉⬇️ @VISxAI @ieeevis #ieeevis #xai
Javier Rando@javirandor
🧐 How is Real-World Gender Bias Reflected in Language Models? We present an interactive article to explore how bias in Language Models aligns with real-world demographics. 🧵👇🏻 Joint work with @melassady @RSevastjanova and A. Theus for #visxai #ieeevis lm-bias.lingvis.io
English
LingVis.io がリツイート
🧐 How is Real-World Gender Bias Reflected in Language Models?
We present an interactive article to explore how bias in Language Models aligns with real-world demographics. 🧵👇🏻
Joint work with @melassady @RSevastjanova and A. Theus for #visxai #ieeevis
lm-bias.lingvis.io
English
LingVis.io がリツイート
To "understand language", masked #languagemodels should capture semantic constraints of function words (e.g., negation, coordination, quantifiers). We show that BERT, RoBERTa, and ALBERT don't: arxiv.org/abs/2209.07836 More: function-words.lingvis.io
To appear @coling2022 . @dbvis
arXiv CS-CL@arxiv_cscl
Negation, Coordination, and Quantifiers in Contextualized Language Models arxiv.org/abs/2209.07836
English
LingVis.io がリツイート
Visual Comparison of Language Model Adaptation
deepai.org/publication/vi…
by @RSevastjanova et al. including @ravfogel
#Vector #ArtificialIntelligence
English
LingVis.io がリツイート
Thank you @NLPParrot for sketching our @IJCAIconf paper on the Rationalization Trap in #XAI - This poster design is amazing! It complements the paper very well!
arxiv.org/abs/2207.06897
#IJCAI2022Workshop @RSevastjanova
NLPurr@NLPurr
Here is the complete graphical abstract for the paper. (6/7)
English
LingVis.io がリツイート
As I mentioned before, I am really glad to be able to merge my love for NLP with that of sketchnoting for science communication.
Today's paper is "Beware the Rationalization Trap!" by @RSevastjanova and @manunna_91
(1/7)
English
LingVis.io がリツイート
What a cool summary of our recent paper on challenges that may occur when designing language model explanations! 😍 Thanks, @NLPParrot ! 😀 Paper: arxiv.org/abs/2207.06897 @manunna_91 @dbvis
NLPurr@NLPurr
As I mentioned before, I am really glad to be able to merge my love for NLP with that of sketchnoting for science communication. Today's paper is "Beware the Rationalization Trap!" by @RSevastjanova and @manunna_91 (1/7)
English
LingVis.io がリツイート
Do you know about the Rationalization Trap in #XAI? We present a conceptual model on challenges that can occur when language model explanations differ from our mental model of language.✍️@manunna_91📢@IJCAIconf #IJCAI2022Workshop
Read more about it here: arxiv.org/abs/2207.06897
English
LingVis.io がリツイート
Happening now at #eurovis2022, @RSevastjanova presenting our work on Language Model Explainability (@LingVISio). Check out lmfingerprints.lingvis.io for more information. #NLProc
Rita Sevastjanova@RSevastjanova
We use visual analytics to create insights into properties learned by transformer-based language models. Our scoring functions can be used as an interpretable alternative to probing classifiers😉 Check out: lmfingerprints.lingvis.io @manunna_91 @HannaJSchaefer @dbvis @EuroVisConf
English
LingVis.io がリツイート
We use visual analytics to create insights into properties learned by transformer-based language models. Our scoring functions can be used as an interpretable alternative to probing classifiers😉 Check out: lmfingerprints.lingvis.io @manunna_91 @HannaJSchaefer @dbvis @EuroVisConf
English
LingVis.io がリツイート
Check out bert-vs-gpt2.dbvis.de and also check out embeddings-explained.lingvis.io to take a closer look at the embedding spaces of language models, such as BERT or GPT2. #LingVisio #VISxAI #ieeevis
VISxAI@VISxAI
Next is “Demystifying the Embedding Space of Language Models” by Rebecca Kehlbeck, @RSevastjanova, @spinthil, Tobias Stähle, @manunna_91, @explainerAI, @lingvisio, @ETH_AI_Center #VISxAI #ieeevis bert-vs-gpt2.dbvis.de
English
LingVis.io がリツイート
Looking forward to presenting @LingVISio (lingvis.io) in the upcoming #vis4dh lab talk session. This is a collaborative project with @RSevastjanova @wjentner @FSperrle @dbvis @KonstanzLing #ieeevis #digitalhumanities #NLProc
VIS4DH Workshop@VIS4DH
Following this, we’ll have 6 lab talks, presenting research from Canberra Design Lab, #LASTIG, @C2DH_LU, LingVis, @udigitalmatters and VisHub! After these talks, we will have a Town Hall discussion and workshop format to discuss where we will go with the #vis4dh workshop. (2/2)
English
LingVis.io がリツイート
BERT and GPT-2 use different strategies to encode a word's position in its context. More about it: bert-vs-gpt2.dbvis.de #LanguageModels #BERT #GPT2 @VISxAI @dbvis @manunna_91 @spinthil
GIF
Rita Sevastjanova@RSevastjanova
What do word embeddings from two popular language models - BERT and GPT-2 - have in common, and in which aspects do they differ significantly?🧐 Check out our blogpost bert-vs-gpt2.dbvis.de - a deep look into embedding contextualization.🎉 @VISxAI @dbvis @manunna_91 @spinthil
English
LingVis.io がリツイート
Next is “Demystifying the Embedding Space of Language Models” by Rebecca Kehlbeck, @RSevastjanova, @spinthil, Tobias Stähle, @manunna_91, @explainerAI, and @lingvisio
bert-vs-gpt2.dbvis.de
Deutsch
LingVis.io がリツイート
Next up, a keynote by Andreas Kerren (Linnaeus University, lnu.se/en/staff/andre…) on "Visual Text Analytics: Overview,
State-of-the-Art, and Challenges"
English