Xinyan Velocity Yu
@velocityyu.bsky.social
NLP/ML @usc, @uwcse. she/her.🏹🤖👩🏻🍳. velocitycavalry.github.io. Multilinguality, retrieval, refining LLMs…
Reposted by Xinyan Velocity Yu
To appear @ #ICLR2025! We show that LMs represent semantically-equivalent inputs across languages, modalities, etc. similarly. This shared representation space is structured by the LM's dominant language, which is also relevant to recent phenomena where LMs "think" in Chinese🀄️ in English🔠 contexts
💡We find that models “think” 💭 in English (or in general, their dominant language) when processing distinct non-English or even non-language data types 🤯 like texts in other languages, arithmetic expressions, code, visual inputs, & audio inputs‼️ 🧵⬇️ arxiv.org/abs/2411.04986
January 22, 2025 at 6:10 PM
To appear @ #ICLR2025! We show that LMs represent semantically-equivalent inputs across languages, modalities, etc. similarly. This shared representation space is structured by the LM's dominant language, which is also relevant to recent phenomena where LMs "think" in Chinese🀄️ in English🔠 contexts
Reposted by Xinyan Velocity Yu
We have released our code at github.com/ZhaofengWu/s.... We hope that this could be useful for future studies understanding the how LMs work!
💡We find that models “think” 💭 in English (or in general, their dominant language) when processing distinct non-English or even non-language data types 🤯 like texts in other languages, arithmetic expressions, code, visual inputs, & audio inputs‼️ 🧵⬇️ arxiv.org/abs/2411.04986
December 17, 2024 at 3:26 PM
We have released our code at github.com/ZhaofengWu/s.... We hope that this could be useful for future studies understanding the how LMs work!
Reposted by Xinyan Velocity Yu
💡We find that models “think” 💭 in English (or in general, their dominant language) when processing distinct non-English or even non-language data types 🤯 like texts in other languages, arithmetic expressions, code, visual inputs, & audio inputs‼️ 🧵⬇️ arxiv.org/abs/2411.04986
December 2, 2024 at 6:08 PM
💡We find that models “think” 💭 in English (or in general, their dominant language) when processing distinct non-English or even non-language data types 🤯 like texts in other languages, arithmetic expressions, code, visual inputs, & audio inputs‼️ 🧵⬇️ arxiv.org/abs/2411.04986