Emmy Liu (@_emliu) 's Twitter Profile
Emmy Liu

@_emliu

PhD student @LTIatCMU, working with @gneubig on NLP || intern @AIatMeta || UofT ‘21 🇨🇦 ||🤖✨🔡

ID: 1461819853430566926

linkhttp://nightingal3.github.io calendar_today19-11-2021 22:13:33

273 Tweet

1,1K Takipçi

499 Takip Edilen

Kayo Yin (@kayo_yin) 's Twitter Profile Photo

Induction heads are commonly associated with in-context learning, but are they the primary driver of ICL at scale? We find that recently discovered "function vector" heads, which encode the ICL task, are the actual primary drivers of few-shot ICL. arxiv.org/abs/2502.14010 🧵

Induction heads are commonly associated with in-context learning, but are they the primary driver of ICL at scale?

We find that recently discovered "function vector" heads, which encode the ICL task, are the actual primary drivers of few-shot ICL.

arxiv.org/abs/2502.14010
🧵
Belinda Li @ ICLR 2025 (@belindazli) 's Twitter Profile Photo

Past work has shown that world state is linearly decodable from LMs trained on text and games like Othello. But how do LMs *compute* these states? We investigate state tracking using permutation composition as a model problem, and discover interpretable, controllable procedures🧵

Past work has shown that world state is linearly decodable from LMs trained on text and games like Othello. But how do LMs *compute* these states? We investigate state tracking using permutation composition as a model problem, and discover interpretable, controllable procedures🧵
Emmy Liu (@_emliu) 's Twitter Profile Photo

Join us this week at the #NAACL student research workshop for an exciting keynote presentation by Philip Resnik! Fun fact, he was a co-creator of the SRW back in the day, so this is coming back full circle ⭐ ⏱️ May 1st, 4-5PM MDT 📍 San Miguel, or online

Join us this week at the #NAACL student research workshop for an exciting keynote presentation by <a href="/psresnik/">Philip Resnik</a>! Fun fact, he was a co-creator of the SRW back in the day, so this is coming back full circle ⭐ 

⏱️ May 1st, 4-5PM MDT 
📍 San Miguel, or online
Emmy Liu (@_emliu) 's Twitter Profile Photo

Returning from #NAACL2025 & had some interesting discussions! One topic that came up a lot was AI scientists and how they should be implemented, evaluated, etc. Used this as inspiration to finish up a blog post on AI science and the state of reviewing: nightingal3.github.io/blog/2025/04/2…

Graham Neubig (@gneubig) 's Twitter Profile Photo

I do think that AI has a lot of promise for science, but we need lots of serious work to get there! One thing I'm very interested in is how we can build AI systems that can effectively judge the quality of research.