Professor Freda Shi in Waterloo's Davis Centre

Freda Shi and her UBC colleagues have won an Outstanding Paper Award at EMNLP 2025. Their work introduces a new benchmark that evaluates how effectively large language models perform meta-linguistic reasoning — using LLMs to better understand languages themselves.

cs.uwaterloo.ca/news/freda-s...

A section called Language Categorization by Geography (“by language family”), which falsely groups languages based on geographic proximity. Arabic, Afrikaans and Swahili are grouped together.

AI people are reclaiming unimportant linguistic terms

Shh, LLM Reasoning reviewers are review-time scaling🫢

Annual Celebration Lūnāris, or better known as ACL🥳

Nearly all eateries near my Beijing home are closed, except that a third McDonald's in the community is opening soon🤯.

Happy xiaonian (Little New Year; en.wikipedia.org/wiki/Little_...) to those who celebrate it! :) Don't forget to eat some candy 🍬 to “bribe” zaowangye (Kitchen God) to say sweet things about you to the Jade Emperor!

NyTorch🗽🌇 #NYC

Thanks for your effort! Everything is working now😌

Today GitHub Copilot notified me I'm running out of quota and have to wait *until* 4172 BC to reset my quota.

Trolley Problem (ARR's Version)

#rant

#ARR AC: That paper is indeed irrelevant. Read this paper first.
Me: But your paper is even not about language models?

Today I first noticed GitHub Copilot coining new words in real life. It suggested "brev-ier" when I wanted to say "more brevity".

#rant
#ARR Reviewer: Read this paper first.
Me: Read my paper first. Or even better, use a better LLM.

#rant What is even more unfair than the reviews is that my pandas asked me for 8PiB of memory.

(Pretending to be a small language model with a bad tokenizer and no positional embedding.)

Orld, low hel!