RecNet
AboutHelpAll Users
Training Language Models with Language Feedback at ScaleJeremy Scheurer, Jon Ander Campos, Tomasz Korbak, Jun Shern Chan, Angelica Chen, Kyunghyun Cho, Ethan Perez
Apr, 2023
Read
M. Omer Gul recommended on on 1/30/2024
The authors propose a simple process for training a model with natural language feedback: Generate multiple refinements on model output by conditioning on the feedback -> Pick the output that best incorporates the feedback -> Finetune model on the refined example.
The Cringe Loss: Learning what language not to modelLeonard Adolphs, Tianyu Gao, Jing Xu, Kurt Shuster, Sainbayar Sukhbaatar, Jason Weston
2023
Read
M. Omer Gul recommended on on 1/9/2024
Neat hacky approach to learn from un-preferred examples in the absence of preference pairs. Rather than simply reduce per-token probabilities, the authors sample a token from the top-k candidates and contrast its probability against that of the undesired token.
Prediction During Language Comprehension: What Is Next?Rachel Ryskin, Mante S. Nieuwland
Sep, 2023
Read
M. Omer Gul recommended on on 1/2/2024
Handy review article on how language production plays a role during language comprehension.
Curriculum Learning with Infant Egocentric VideosSaber Sheybani, Himanshu Hansaria, Justin N. Wood, Linda B. Smith, Zoran Tiganj
2023
Read
M. Omer Gul recommended on on 12/5/2023
The authors explore whether self-supervised learning on infant ego-centric videos can result in good representations, why, and whether performing curriculum learning based on developmental stage helps perform better vs other data regimes.
Communication Breakdown: On the low mutual intelligibility between human and neural captioningRoberto Dessì, Eleonora Gualdoni, Francesca Franzon, Gemma Boleda, Marco Baroni
2022
Read
M. Omer Gul recommended on on 11/28/2023
A simple paper demonstrating the difference in model retrieval performance when given human and model generated captions. Models perform better with model captions, even though humans cannot necessarily make sense of the same captions.
The Generative AI Paradox: "What It Can Create, It May Not Understand"Peter West, Ximing Lu, Nouha Dziri, Faeze Brahman, Linjie Li, Jena D. Hwang, Liwei Jiang, Jillian Fisher, Abhilasha Ravichander, Khyathi Raghavi Chandu, Benjamin Newman, Pang Wei Koh, Allyson Ettinger, Yejin Choi
Nov, 2023
Read
M. Omer Gul recommended on on 11/21/2023
The authors probe an observed discrepancy between models' performance in generation tasks and those tasks' discriminative variants. A different angle on the relationship between comprehension and production than my project.
© 2024 RecNet. All rights reserved.