Eric M. Smith
YOU?
Author Swipe
View article: The Alignment Waltz: Jointly Training Agents to Collaborate for Safety
The Alignment Waltz: Jointly Training Agents to Collaborate for Safety Open
Harnessing the power of LLMs requires a delicate dance between being helpful and harmless. This creates a fundamental tension between two competing challenges: vulnerability to adversarial attacks that elicit unsafe content, and a tendency…
View article: Llama Guard 3-1B-INT4: Compact and Efficient Safeguard for Human-AI Conversations
Llama Guard 3-1B-INT4: Compact and Efficient Safeguard for Human-AI Conversations Open
This paper presents Llama Guard 3-1B-INT4, a compact and efficient Llama Guard model, which has been open-sourced to the community during Meta Connect 2024. We demonstrate that Llama Guard 3-1B-INT4 can be deployed on resource-constrained …
View article: Llama Guard 3 Vision: Safeguarding Human-AI Image Understanding Conversations
Llama Guard 3 Vision: Safeguarding Human-AI Image Understanding Conversations Open
We introduce Llama Guard 3 Vision, a multimodal LLM-based safeguard for human-AI conversations that involves image understanding: it can be used to safeguard content for both multimodal LLM inputs (prompt classification) and outputs (respo…
View article: Scale invariance in early embryonic development
Scale invariance in early embryonic development Open
The expression of a few key genes determines the body plan of the fruit fly. We show that the spatial expression patterns for several of these genes scale precisely with embryo size. Discrete positional markers such as the peaks in striped…
View article: The Root Shapes the Fruit: On the Persistence of Gender-Exclusive Harms in Aligned Language Models
The Root Shapes the Fruit: On the Persistence of Gender-Exclusive Harms in Aligned Language Models Open
Natural-language assistants are designed to provide users with helpful responses while avoiding harmful outputs, largely achieved through alignment to human preferences. Yet there is limited understanding of whether alignment techniques ma…
View article: Persistent Pre-Training Poisoning of LLMs
Persistent Pre-Training Poisoning of LLMs Open
Large language models are pre-trained on uncurated text datasets consisting of trillions of tokens scraped from the Web. Prior work has shown that: (1) web-scraped pre-training datasets can be practically poisoned by malicious actors; and …
View article: Backtracking Improves Generation Safety
Backtracking Improves Generation Safety Open
Text generation has a fundamental limitation almost by definition: there is no taking back tokens that have been generated, even when they are clearly problematic. In the context of language model safety, when a partial unsafe generation i…
View article: Towards Safety and Helpfulness Balanced Responses via Controllable Large Language Models
Towards Safety and Helpfulness Balanced Responses via Controllable Large Language Models Open
As large language models (LLMs) become easily accessible nowadays, the trade-off between safety and helpfulness can significantly impact user experience. A model that prioritizes safety will cause users to feel less engaged and assisted wh…
View article: Scale invariance in early embryonic development
Scale invariance in early embryonic development Open
The body plan of the fruit fly is determined by the expression of just a handful of genes. We show that the spatial patterns of expression for several of these genes scale precisely with the size of the embryo. Concretely, discrete positio…
View article: ROBBIE: Robust Bias Evaluation of Large Generative Language Models
ROBBIE: Robust Bias Evaluation of Large Generative Language Models Open
As generative large language models (LLMs) grow more performant and prevalent, we must develop comprehensive enough tools to measure and improve their fairness. Different prompt-based datasets can be used to measure social bias across mult…
View article: The Gender-GAP Pipeline: A Gender-Aware Polyglot Pipeline for Gender Characterisation in 55 Languages
The Gender-GAP Pipeline: A Gender-Aware Polyglot Pipeline for Gender Characterisation in 55 Languages Open
Gender biases in language generation systems are challenging to mitigate. One possible source for these biases is gender representation disparities in the training and evaluation data. Despite recent progress in documenting this problem an…
View article: In Memoriam: Robert A. “Bob” Jones
In Memoriam: Robert A. “Bob” Jones Open
August 3, 1930 – March 31, 2023 Robert A. “Bob” Jones of South Windsor, Connecticut passed away at age 92 on March 31, 2023. Bob was a ground-breaking fisheries biologist who had impact at the state, interstate, and international levels. H…
View article: Llama 2: Open Foundation and Fine-Tuned Chat Models
Llama 2: Open Foundation and Fine-Tuned Chat Models Open
In this work, we develop and release Llama 2, a collection of pretrained and fine-tuned large language models (LLMs) ranging in scale from 7 billion to 70 billion parameters. Our fine-tuned LLMs, called Llama 2-Chat, are optimized for dial…
View article: A nondepleting anti-CD19 antibody impairs B cell function and inhibits autoimmune diseases
A nondepleting anti-CD19 antibody impairs B cell function and inhibits autoimmune diseases Open
B cells contribute to multiple aspects of autoimmune disorders, and B cell-targeting therapies, including B cell depletion, have been proven to be efficacious in treatment of multiple autoimmune diseases. However, the development of novel …
View article: Improving Open Language Models by Learning from Organic Interactions
Improving Open Language Models by Learning from Organic Interactions Open
We present BlenderBot 3x, an update on the conversational model BlenderBot 3, which is now trained using organic conversation and feedback data from participating users of the system in order to improve both its skills and safety. We are p…
View article: The Gender-GAP Pipeline: A Gender-Aware Polyglot Pipeline for Gender Characterisation in 55 Languages
The Gender-GAP Pipeline: A Gender-Aware Polyglot Pipeline for Gender Characterisation in 55 Languages Open
Gender biases in language generation systems are challenging to mitigate.One possible source for these biases is gender representation disparities in the training and evaluation data.Despite recent progress in documenting this problem and …
View article: ROBBIE: Robust Bias Evaluation of Large Generative Language Models
ROBBIE: Robust Bias Evaluation of Large Generative Language Models Open
David Esiobu, Xiaoqing Tan, Saghar Hosseini, Megan Ung, Yuchen Zhang, Jude Fernandes, Jane Dwivedi-Yu, Eleonora Presani, Adina Williams, Eric Smith. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 20…
View article: BlenderBot 3: a deployed conversational agent that continually learns to responsibly engage
BlenderBot 3: a deployed conversational agent that continually learns to responsibly engage Open
We present BlenderBot 3, a 175B parameter dialogue model capable of open-domain conversation with access to the internet and a long-term memory, and having been trained on a large number of user defined tasks. We release both the model wei…
View article: "I'm sorry to hear that": Finding New Biases in Language Models with a Holistic Descriptor Dataset
"I'm sorry to hear that": Finding New Biases in Language Models with a Holistic Descriptor Dataset Open
As language models grow in popularity, it becomes increasingly important to clearly measure all possible markers of demographic identity in order to avoid perpetuating existing societal harms. Many datasets for measuring bias currently exi…
View article: Human Evaluation of Conversations is an Open Problem: comparing the sensitivity of various methods for evaluating dialogue agents
Human Evaluation of Conversations is an Open Problem: comparing the sensitivity of various methods for evaluating dialogue agents Open
At the heart of improving conversational AI is the open problem of how to evaluate conversations. Issues with automatic metrics are well known (Liu et al., 2016, arXiv:1603.08023), with human evaluations still considered the gold standard.…
View article: “I’m sorry to hear that”: Finding New Biases in Language Models with a Holistic Descriptor Dataset
“I’m sorry to hear that”: Finding New Biases in Language Models with a Holistic Descriptor Dataset Open
As language models grow in popularity, it becomes increasingly important to clearly measure all possible markers of demographic identity in order to avoid perpetuating existing societal harms. Many datasets for measuring bias currently exi…
View article: Perturbation Augmentation for Fairer NLP
Perturbation Augmentation for Fairer NLP Open
Unwanted and often harmful social biases are becoming ever more salient in NLP research, affecting both models and datasets. In this work, we ask whether training on demographically perturbed data leads to fairer language models. We collec…
View article: Hi, my name is Martha: Using names to measure and mitigate bias in generative dialogue models
Hi, my name is Martha: Using names to measure and mitigate bias in generative dialogue models Open
All AI models are susceptible to learning biases in data that they are trained on. For generative dialogue models, being trained on real human conversations containing unbalanced gender and race/ethnicity references can lead to models that…
View article: Basic residues at the C-gate of DNA gyrase are involved in DNA supercoiling
Basic residues at the C-gate of DNA gyrase are involved in DNA supercoiling Open
DNA gyrase is a type II topoisomerase that is responsible for maintaining the topological state of bacterial and some archaeal genomes. It uses an ATP-dependent two-gate strand-passage mechanism that is shared among all type II topoisomera…
View article: Multi-Modal Open-Domain Dialogue
Multi-Modal Open-Domain Dialogue Open
Recent work in open-domain conversational agents has demonstrated that significant improvements in model engagingness and humanness metrics can be achieved via massive scaling in both pre-training data and model size (Adiwardana et al., 20…
View article: Recipes for Building an Open-Domain Chatbot
Recipes for Building an Open-Domain Chatbot Open
Building open-domain chatbots is a challenging area for machine learning research. While prior work has shown that scaling neural models in the number of parameters and the size of the data they are trained on gives improved results, we sh…
View article: Controlling Style in Generated Dialogue
Controlling Style in Generated Dialogue Open
Open-domain conversation models have become good at generating natural-sounding dialogue, using very large architectures with billions of trainable parameters. The vast training data required to train these architectures aggregates many di…
View article: Open-Domain Conversational Agents: Current Progress, Open Problems, and Future Directions
Open-Domain Conversational Agents: Current Progress, Open Problems, and Future Directions Open
We present our view of what is necessary to build an engaging open-domain conversational agent: covering the qualities of such an agent, the pieces of the puzzle that have been built so far, and the gaping holes we have not filled yet. We …
View article: Can You Put it All Together: Evaluating Conversational Agents' Ability\n to Blend Skills
Can You Put it All Together: Evaluating Conversational Agents' Ability\n to Blend Skills Open
Being engaging, knowledgeable, and empathetic are all desirable general\nqualities in a conversational agent. Previous work has introduced tasks and\ndatasets that aim to help agents to learn those qualities in isolation and\ngauge how wel…