Papers by Max Bartolo
No Need for Explanations: LLMs can implicitly learn from mistakes in-context (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing literature assumes that correct answers to large language models must be accompanied by comprehensive rationales to be helpful. |
| Approach: | They propose to show incorrect answers to Large Language Models (LLMs) as a popular strategy to improve their performance in reasoning-intensive tasks. |
| Outcome: | The proposed approach outperforms chain-of-thought prompting in math reasoning tasks. |
Dynatask: A Framework for Creating Dynamic AI Benchmark Tasks (2022.acl-demo)
Copied to clipboard
Tristan Thrush, Kushal Tirumala, Anmol Gupta, Max Bartolo, Pedro Rodriguez, Tariq Kane, William Gaviria Rojas, Peter Mattson, Adina Williams, Douwe Kiela
| Challenge: | Open source system for setting up custom NLP tasks aims to lower technical knowledge and effort required for hosting and evaluating state-of-the-art models. |
| Approach: | They propose to integrate Dynatask with Dynabench to simplify benchmarking . they use a dataset to collect and clean data and train and evaluate models . |
| Outcome: | Dynatask is an open source system for setting up custom NLP tasks . it is integrated with Dynabench, a research platform for rethinking benchmarking in AI . |
Interpretation of Natural Language Rules in Conversational Machine Reading (D18-1)
Copied to clipboard
Marzieh Saeidi, Max Bartolo, Patrick Lewis, Sameer Singh, Tim Rocktäschel, Mike Sheldon, Guillaume Bouchard, Sebastian Riedel
| Challenge: | Existing work on question answering problems requires the reading of text because it contains a recipe to derive an answer together with the reader’s background knowledge. |
| Approach: | They formalise a task and develop a crowd-sourcing strategy to collect 37k task instances based on real-world rules and crowd-generated questions and scenarios. |
| Outcome: | The proposed task is based on 37k task instances based in real-world rules and crowd-generated questions and scenarios. |
Fishing for Magikarp: Automatically Detecting Under-trained Tokens in Large Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Disconnect between tokenizer creation and model training in language models allows for specific inputs, such as the infamous SolidGoldMagikarp token, to induce unwanted model behaviour. |
| Approach: | They propose to use tokenizer analysis, model weight-based indicators, and prompting techniques to detect problematic tokens in large language models. |
| Outcome: | The proposed methods show that tokenizers are under-trained across a diverse set of models and provide insights into improving the efficiency and safety of language models. |
Fantastically Ordered Prompts and Where to Find Them: Overcoming Few-Shot Prompt Order Sensitivity (2022.acl-long)
Copied to clipboard
| Challenge: | Large pretrained language models can generate text classification results that match fully supervised models. |
| Approach: | They propose to use a few sample training to determine which permutations are performant . they use generative language models to construct an artificial development set . |
| Outcome: | The proposed model outperforms fully-supervised models in eleven text classification tasks. |
Improving Question Answering Model Robustness with Synthetic Adversarial Data Generation (2021.emnlp-main)
Copied to clipboard
| Challenge: | a new approach to generate adversarial data is needed to improve question answering models . crowdworkers can fool a model only 8.8% of the time, compared to 17.6% for a trained model without synthetic data. |
| Approach: | They develop a pipeline that generates questions and then filters or labels them to improve quality. |
| Outcome: | The proposed approach improves state-of-the-art on a human-written adversarial dataset by 3.7F1 and improves model generalisation on nine of the twelve MRQA datasets. |
Dynabench: Rethinking Benchmarking in NLP (2021.naacl-main)
Copied to clipboard
Douwe Kiela, Max Bartolo, Yixin Nie, Divyansh Kaushik, Atticus Geiger, Zhengxuan Wu, Bertie Vidgen, Grusha Prasad, Amanpreet Singh, Pratik Ringshia, Zhiyi Ma, Tristan Thrush, Sebastian Riedel, Zeerak Waseem, Pontus Stenetorp, Robin Jia, Mohit Bansal, Christopher Potts, Adina Williams
| Challenge: | Dynabench is an open-source platform for dynamic dataset creation and model benchmarking. |
| Approach: | They propose an open-source platform for dynamic dataset creation and model benchmarking. |
| Outcome: | The proposed platform can be used to create models that fail on simple challenges and falter in real-world scenarios. |
Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension (2020.tacl-1)
Copied to clipboard
| Challenge: | Innovations in annotation methodologies have been a catalyst for Reading Comprehension (RC) datasets and models. |
| Approach: | They propose to use a model-in-the-annotation-loop approach to train adversarial models in three different settings to explore reproducibility of the adversarial effect, transfer from data collected with varying model- in-the loop strengths, and generalization to data collected without a modeling model. |
| Outcome: | The proposed approach achieves 39.9F1 on questions it cannot answer when trained on SQUAD, but lower than when trained using RoBERTa itself (41.0F1). |
Aya Dataset: An Open-Access Collection for Multilingual Instruction Tuning (2024.acl-long)
Copied to clipboard
Shivalika Singh, Freddie Vargus, Daniel D’souza, Börje Karlsson, Abinaya Mahendiran, Wei-Yin Ko, Herumb Shandilya, Jay Patel, Deividas Mataciunas, Laura O’Mahony, Mike Zhang, Ramith Hettiarachchi, Joseph Wilson, Marina Machado, Luisa Moura, Dominik Krzemiński, Hakimeh Fadaei, Irem Ergun, Ifeoma Okoh, Aisha Alaagib, Oshan Mudannayake, Zaid Alyafeai, Vu Chien, Sebastian Ruder, Surya Guthikonda, Emad Alghamdi, Sebastian Gehrmann, Niklas Muennighoff, Max Bartolo, Julia Kreutzer, Ahmet Üstün, Marzieh Fadaee, Sara Hooker
| Challenge: | Existing datasets in the English language are mostly in the realm of instruction fine-tuning . aya dataset, the Aya Collection, and the AYa Evaluation Suite are key resources . |
| Approach: | They aim to build a human-curated instruction-following dataset spanning 65 languages . they work with fluent speakers of languages from around the world to collect natural instances of instructions and completions . |
| Outcome: | The goal is to build a human-curated instruction-following dataset spanning 65 languages. |
Contrasting Human- and Machine-Generated Word-Level Adversarial Examples for Text Classification (2021.emnlp-main)
Copied to clipboard
| Challenge: | Recent work has raised the question of whether valid adversarial inputs are feasible. |
| Approach: | They analyze how human-generated adversarial examples compare to the best algorithms . they use crowdsourcing to modify words in an input text with immediate feedback . |
| Outcome: | The proposed algorithms are not more efficient than the best to generate natural-reading, sentiment-preserving examples. |
Models in the Loop: Aiding Crowdworkers with Generative Annotation Assistants (2022.naacl-main)
Copied to clipboard
| Challenge: | Dynamic Adversarial Data Collection (DADC) is a time-consuming and costly approach . DADC is based on training data collected from adversarial and out-of-domain settings . |
| Approach: | They propose a dynamic data collection approach that uses generator-in-the-loop models to provide real-time suggestions that annotators can approve, modify, or reject. |
| Outcome: | The proposed model is more robust in adversarial and out-of-domain settings and harder for humans to fool. |
Improving Reward Models with Synthetic Critiques (2025.findings-naacl)
Copied to clipboard
| Challenge: | a recent study shows that reward models overfit on superficial features, hindering generalization performance . prevailing approach to training preference-based reward models presents several challenges . |
| Approach: | They propose a method that uses synthetic natural language critiques to provide additional feedback to large language models. |
| Outcome: | The proposed approach improves performance and data efficiency of RMs initialized from different pretrained models, reducing the reliance on costly human annotations. |
Undersensitivity in Neural Reading Comprehension (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Existing models generalise well to in-distribution test sets, yet perform poorly on adversarially selected data. |
| Approach: | They propose an adversarial attack which searches among semantic variations of the question for which a model erroneously predicts the same answer, and with even higher probability. |
| Outcome: | The proposed attack reduces the vulnerability of models trained on SQuAD2.0 and NewsQA, and outperforms a conventional model by as much as 10.9% F1. |