Taylor Sorensen
@taylor-sorensen.bsky.social
NLP PhD Candidate at UW
Did you know that LLMs suffer from serious mode collapse?
For example, if you ask models to tell you a joke, they almost always tell you the same joke? This is true across samples and even across model families!
Why does this happen? Can we improve it?
For example, if you ask models to tell you a joke, they almost always tell you the same joke? This is true across samples and even across model families!
Why does this happen? Can we improve it?
October 8, 2025 at 2:22 PM
Did you know that LLMs suffer from serious mode collapse?
For example, if you ask models to tell you a joke, they almost always tell you the same joke? This is true across samples and even across model families!
Why does this happen? Can we improve it?
For example, if you ask models to tell you a joke, they almost always tell you the same joke? This is true across samples and even across model families!
Why does this happen? Can we improve it?
Reposted by Taylor Sorensen
Want to know what training data has been memorized by models like GPT-4?
We propose information-guided probes, a method to uncover memorization evidence in *completely black-box* models,
without requiring access to
🙅♀️ Model weights
🙅♀️ Training data
🙅♀️ Token probabilities 🧵 (1/5)
We propose information-guided probes, a method to uncover memorization evidence in *completely black-box* models,
without requiring access to
🙅♀️ Model weights
🙅♀️ Training data
🙅♀️ Token probabilities 🧵 (1/5)
Information-Guided Identification of Training Data Imprint in (Proprietary) Large Language Models
High-quality training data has proven crucial for developing performant large language models (LLMs). However, commercial LLM providers disclose few, if any, details about the data used for training. ...
arxiv.org
March 21, 2025 at 7:08 PM
Want to know what training data has been memorized by models like GPT-4?
We propose information-guided probes, a method to uncover memorization evidence in *completely black-box* models,
without requiring access to
🙅♀️ Model weights
🙅♀️ Training data
🙅♀️ Token probabilities 🧵 (1/5)
We propose information-guided probes, a method to uncover memorization evidence in *completely black-box* models,
without requiring access to
🙅♀️ Model weights
🙅♀️ Training data
🙅♀️ Token probabilities 🧵 (1/5)
🤔🤖Most AI systems assume there’s just one right answer—but many tasks have reasonable disagreement. How can we better model human variation? 🌍✨
We propose modeling at the individual-level using open-ended, textual value profiles! 🗣️📝
arxiv.org/abs/2503.15484
We propose modeling at the individual-level using open-ended, textual value profiles! 🗣️📝
arxiv.org/abs/2503.15484
March 20, 2025 at 3:45 AM
🤔🤖Most AI systems assume there’s just one right answer—but many tasks have reasonable disagreement. How can we better model human variation? 🌍✨
We propose modeling at the individual-level using open-ended, textual value profiles! 🗣️📝
arxiv.org/abs/2503.15484
We propose modeling at the individual-level using open-ended, textual value profiles! 🗣️📝
arxiv.org/abs/2503.15484
I’ll be in Vancouver this weekend for the NeurIPS workshops (go pluralistic alignment!)
DMs are open if anyone wants to chat! :)
DMs are open if anyone wants to chat! :)
December 13, 2024 at 9:36 AM
I’ll be in Vancouver this weekend for the NeurIPS workshops (go pluralistic alignment!)
DMs are open if anyone wants to chat! :)
DMs are open if anyone wants to chat! :)
Reposted by Taylor Sorensen
🌟 Come to our Pluralistic Alignment Workshop at #NeurIPS2024!
🗓️ December 14
📍 West Meeting Room 116, 117
Join us to explore pluralistic perspectives in alignment with an incredible lineup of talks and speakers!
🔗 Full schedule & details: pluralistic-alignment.github.io
🗓️ December 14
📍 West Meeting Room 116, 117
Join us to explore pluralistic perspectives in alignment with an incredible lineup of talks and speakers!
🔗 Full schedule & details: pluralistic-alignment.github.io
December 6, 2024 at 1:04 AM
🌟 Come to our Pluralistic Alignment Workshop at #NeurIPS2024!
🗓️ December 14
📍 West Meeting Room 116, 117
Join us to explore pluralistic perspectives in alignment with an incredible lineup of talks and speakers!
🔗 Full schedule & details: pluralistic-alignment.github.io
🗓️ December 14
📍 West Meeting Room 116, 117
Join us to explore pluralistic perspectives in alignment with an incredible lineup of talks and speakers!
🔗 Full schedule & details: pluralistic-alignment.github.io
Reposted by Taylor Sorensen
If you are headed to NeurIPS, please join for our Pluralistic Alignment workshop! We have a great set of speakers from a range of backgrounds. & all the papers that will be presented at the workshop are posted: pluralistic-alignment.github.io lmk if you'd like to catch up at the conference too! :)
Pluralistic Alignment @ NeurIPS 2024
Pluralistic Alignment
pluralistic-alignment.github.io
December 5, 2024 at 2:56 AM
If you are headed to NeurIPS, please join for our Pluralistic Alignment workshop! We have a great set of speakers from a range of backgrounds. & all the papers that will be presented at the workshop are posted: pluralistic-alignment.github.io lmk if you'd like to catch up at the conference too! :)