1 code implementation • 15 Apr 2024 • Usman Anwar, Abulhair Saparov, Javier Rando, Daniel Paleka, Miles Turpin, Peter Hase, Ekdeep Singh Lubana, Erik Jenner, Stephen Casper, Oliver Sourbut, Benjamin L. Edelman, Zhaowei Zhang, Mario Günther, Anton Korinek, Jose Hernandez-Orallo, Lewis Hammond, Eric Bigelow, Alexander Pan, Lauro Langosco, Tomasz Korbak, Heidi Zhang, Ruiqi Zhong, Seán Ó hÉigeartaigh, Gabriel Recchia, Giulio Corsi, Alan Chan, Markus Anderljung, Lilian Edwards, Yoshua Bengio, Danqi Chen, Samuel Albanie, Tegan Maharaj, Jakob Foerster, Florian Tramer, He He, Atoosa Kasirzadeh, Yejin Choi, David Krueger
This work identifies 18 foundational challenges in assuring the alignment and safety of large language models (LLMs).
no code implementations • 13 Feb 2024 • Sijia Liu, Yuanshun Yao, Jinghan Jia, Stephen Casper, Nathalie Baracaldo, Peter Hase, Xiaojun Xu, Yuguang Yao, Hang Li, Kush R. Varshney, Mohit Bansal, Sanmi Koyejo, Yang Liu
We explore machine unlearning (MU) in the domain of large language models (LLMs), referred to as LLM unlearning.
1 code implementation • 12 Jan 2024 • Peter Hase, Mohit Bansal, Peter Clark, Sarah Wiegreffe
In this paper, we present the surprising conclusion that current language models often generalize relatively well from easy to hard data, even performing as well as "oracle" models trained on hard data.
1 code implementation • 29 Sep 2023 • Vaidehi Patil, Peter Hase, Mohit Bansal
Experimentally, we show that even state-of-the-art model editing methods such as ROME struggle to truly delete factual information from models like GPT-J, as our whitebox and blackbox attacks can recover "deleted" information from an edited model 38% of the time.
no code implementations • 27 Jul 2023 • Stephen Casper, Xander Davies, Claudia Shi, Thomas Krendl Gilbert, Jérémy Scheurer, Javier Rando, Rachel Freedman, Tomasz Korbak, David Lindner, Pedro Freire, Tony Wang, Samuel Marks, Charbel-Raphaël Segerie, Micah Carroll, Andi Peng, Phillip Christoffersen, Mehul Damani, Stewart Slocum, Usman Anwar, Anand Siththaranjan, Max Nadeau, Eric J. Michaud, Jacob Pfau, Dmitrii Krasheninnikov, Xin Chen, Lauro Langosco, Peter Hase, Erdem Biyik, Anca Dragan, David Krueger, Dorsa Sadigh, Dylan Hadfield-Menell
Reinforcement learning from human feedback (RLHF) is a technique for training AI systems to align with human goals.
1 code implementation • 15 Jun 2023 • Swarnadeep Saha, Peter Hase, Mohit Bansal
We first show that teacher LLMs can indeed intervene on student reasoning to improve their performance.
1 code implementation • NeurIPS 2023 • Peter Hase, Mohit Bansal, Been Kim, Asma Ghandeharioun
This finding raises questions about how past work relies on Causal Tracing to select which model layers to edit.
1 code implementation • 14 Nov 2022 • Swarnadeep Saha, Peter Hase, Nazneen Rajani, Mohit Bansal
We observe that (1) GPT-3 explanations are as grammatical as human explanations regardless of the hardness of the test samples, (2) for easy examples, GPT-3 generates highly supportive explanations but human explanations are more generalizable, and (3) for hard examples, human explanations are significantly better than GPT-3 explanations both in terms of label-supportiveness and generalizability judgements.
1 code implementation • 21 Sep 2022 • Swarnadeep Saha, Shiyue Zhang, Peter Hase, Mohit Bansal
We demonstrate that SP-Search effectively represents the generative process behind human summaries using modules that are typically faithful to their intended behavior.
1 code implementation • 22 Jun 2022 • Zhuofan Ying, Peter Hase, Mohit Bansal
In this paper, we show that model FI supervision can meaningfully improve VQA model accuracy as well as performance on several Right-for-the-Right-Reason (RRR) metrics by optimizing for four key model objectives: (1) accurate predictions given limited but sufficient information (Sufficiency); (2) max-entropy predictions given no important information (Uncertainty); (3) invariance of predictions to changes in unimportant features (Invariance); and (4) alignment between model FI explanations and human FI explanations (Plausibility).
2 code implementations • 14 Mar 2022 • Archiki Prasad, Peter Hase, Xiang Zhou, Mohit Bansal
Providing natural language instructions in prompts is a useful new paradigm for improving task performance of large language models in a zero-shot setting.
1 code implementation • 26 Nov 2021 • Peter Hase, Mona Diab, Asli Celikyilmaz, Xian Li, Zornitsa Kozareva, Veselin Stoyanov, Mohit Bansal, Srinivasan Iyer
In this paper, we discuss approaches to detecting when models have beliefs about the world, and we improve on methods for updating model beliefs to be more truthful, with a focus on methods based on learned optimizers or hypernetworks.
1 code implementation • 1 Nov 2021 • Prateek Yadav, Peter Hase, Mohit Bansal
Current approaches try to optimize for the cost incurred by users when adopting a recourse, but they assume that all users share the same cost function.
1 code implementation • NeurIPS 2021 • Peter Hase, Harry Xie, Mohit Bansal
In this paper, we study several under-explored dimensions of FI explanations, providing conceptual and empirical improvements for this form of explanation.
1 code implementation • LNLS (ACL) 2022 • Peter Hase, Mohit Bansal
In order to carefully control important properties of the data and explanations, we introduce a synthetic dataset for experiments, and we also make use of three existing datasets with explanations: e-SNLI, TACRED, and SemEval.
1 code implementation • EMNLP 2021 • Han Guo, Nazneen Fatema Rajani, Peter Hase, Mohit Bansal, Caiming Xiong
With the availability of the fast influence functions, we demonstrate their usefulness in four applications.
1 code implementation • Findings of the Association for Computational Linguistics 2020 • Peter Hase, Shiyue Zhang, Harry Xie, Mohit Bansal
We provide code for the experiments in this paper at https://github. com/peterbhase/LAS-NL-Explanations
1 code implementation • ACL 2020 • Peter Hase, Mohit Bansal
Through two kinds of simulation tests involving text and tabular data, we evaluate five explanations methods: (1) LIME, (2) Anchor, (3) Decision Boundary, (4) a Prototype model, and (5) a Composite approach that combines explanations from each method.
1 code implementation • 25 Jun 2019 • Peter Hase, Chaofan Chen, Oscar Li, Cynthia Rudin
Hence, we may find distinct explanations for the prediction an image receives at each level of the taxonomy.
2 code implementations • 13 Nov 2018 • John Benhardt, Peter Hase, Liuyi Zhu, Cynthia Rudin
We provide an approach for generating beautiful poetry.