Eric Wallace
YOU?
Author Swipe
View article: Characterizing pain in patients with Fabry disease: findings from a web-based cross-sectional survey in the US
Characterizing pain in patients with Fabry disease: findings from a web-based cross-sectional survey in the US Open
Background Fabry disease (FD) is a rare, progressive disorder caused by pathogenic variants of the GLA gene resulting in the accumulation of toxic metabolites. Pain is a hallmark of FD, and patients often present with heterogeneous pain pr…
View article: Trading Inference-Time Compute for Adversarial Robustness
Trading Inference-Time Compute for Adversarial Robustness Open
We conduct experiments on the impact of increasing inference-time compute in reasoning models (specifically OpenAI o1-preview and o1-mini) on their robustness to adversarial attacks. We find that across a variety of attacks, increased infe…
View article: OpenAI o1 System Card
OpenAI o1 System Card Open
The o1 model series is trained with large-scale reinforcement learning to reason using chain of thought. These advanced reasoning capabilities provide new avenues for improving the safety and robustness of our models. In particular, our mo…
View article: The underdiagnosed kidney burden of Fabry disease in females
The underdiagnosed kidney burden of Fabry disease in females Open
Aim: Progressive kidney deterioration can occur in females with Fabry disease. Despite this, many females go undiagnosed or are labeled “carriers”. Our study aimed to review biopsy findings of adult female Fabry patients with creatinine va…
View article: Predicting Emergent Capabilities by Finetuning
Predicting Emergent Capabilities by Finetuning Open
A fundamental open challenge in modern LLM scaling is the lack of understanding around emergent capabilities. In particular, language model pretraining loss is known to be highly predictable as a function of compute. However, downstream ca…
View article: The Role of Vitamin D Deficiency in Hepatic Encephalopathy: A Review of Pathophysiology, Clinical Outcomes, and Therapeutic Potential
The Role of Vitamin D Deficiency in Hepatic Encephalopathy: A Review of Pathophysiology, Clinical Outcomes, and Therapeutic Potential Open
Hepatic encephalopathy (HE) is a neuropsychiatric condition frequently associated with cirrhosis and portosystemic shunting (PSS). It imposes a significant clinical and economic burden, with increasing attention toward identifying modifiab…
View article: Discoveries and insights from implementing telehealth in a tele-acute unit: A retrospective study
Discoveries and insights from implementing telehealth in a tele-acute unit: A retrospective study Open
Objective: This study examines the impact of telehealth nursing interventions on length of stay (LOS) and ratio of LOS to risk-adjusted length of stay comparing tele-acute and traditional units.Methods: Retrospective data from 6,999 patien…
View article: A phase <scp>III</scp>, open‐label clinical trial evaluating pegunigalsidase alfa administered every 4 weeks in adults with Fabry disease previously treated with other enzyme replacement therapies
A phase <span>III</span>, open‐label clinical trial evaluating pegunigalsidase alfa administered every 4 weeks in adults with Fabry disease previously treated with other enzyme replacement therapies Open
Pegunigalsidase alfa, a PEGylated α‐galactosidase A enzyme replacement therapy (ERT) for Fabry disease, has a longer plasma half‐life than other ERTs administered intravenously every 2 weeks (E2W). BRIGHT (NCT03180840) was a phase III, ope…
View article: Covert Malicious Finetuning: Challenges in Safeguarding LLM Adaptation
Covert Malicious Finetuning: Challenges in Safeguarding LLM Adaptation Open
Black-box finetuning is an emerging interface for adapting state-of-the-art language models to user needs. However, such access may also let malicious actors undermine model safety. To demonstrate the challenge of defending finetuning inte…
View article: Unfamiliar Finetuning Examples Control How Language Models Hallucinate
Unfamiliar Finetuning Examples Control How Language Models Hallucinate Open
Large language models are known to hallucinate when faced with unfamiliar queries, but the underlying mechanism that govern how models hallucinate are not yet fully understood. In this work, we find that unfamiliar examples in the models' …
View article: Characterizing pain in patients with Fabry disease: Findings from a web-based cross-sectional survey in the US
Characterizing pain in patients with Fabry disease: Findings from a web-based cross-sectional survey in the US Open
Background: Fabry disease (FD) is a rare, progressive disorder caused by pathogenic variants of the GLA gene resulting in the accumulation of toxic metabolites. Pain is a hallmark of FD, and patients often present with heterogeneous pain p…
View article: Scalable Extraction of Training Data from (Production) Language Models
Scalable Extraction of Training Data from (Production) Language Models Open
This paper studies extractable memorization: training data that an adversary can efficiently extract by querying a machine learning model without prior knowledge of the training dataset. We show an adversary can extract gigabytes of traini…
View article: Head-to-head trial of pegunigalsidase alfa versus agalsidase beta in patients with Fabry disease and deteriorating renal function: results from the 2-year randomised phase III BALANCE study
Head-to-head trial of pegunigalsidase alfa versus agalsidase beta in patients with Fabry disease and deteriorating renal function: results from the 2-year randomised phase III BALANCE study Open
Background Pegunigalsidase alfa is a PEGylated α-galactosidase A enzyme replacement therapy. BALANCE ( NCT02795676 ) assessed non-inferiority of pegunigalsidase alfa versus agalsidase beta in adults with Fabry disease with an annualised es…
View article: Privacy Side Channels in Machine Learning Systems
Privacy Side Channels in Machine Learning Systems Open
Most current approaches for protecting privacy in machine learning (ML) assume that models exist in a vacuum. Yet, in reality, these models are part of larger systems that include components for training data filtering, output monitoring, …
View article: SILO Language Models: Isolating Legal Risk In a Nonparametric Datastore
SILO Language Models: Isolating Legal Risk In a Nonparametric Datastore Open
The legality of training language models (LMs) on copyrighted or otherwise restricted data is under intense debate. However, as we show, model performance significantly degrades if trained only on low-risk text (e.g., out-of-copyright book…
View article: The False Promise of Imitating Proprietary LLMs
The False Promise of Imitating Proprietary LLMs Open
An emerging method to cheaply improve a weaker language model is to finetune it on outputs from a stronger model, such as a proprietary system like ChatGPT (e.g., Alpaca, Self-Instruct, and others). This approach looks to cheaply imitate t…
View article: Poisoning Language Models During Instruction Tuning
Poisoning Language Models During Instruction Tuning Open
Instruction-tuned LMs such as ChatGPT, FLAN, and InstructGPT are finetuned on datasets that contain user-submitted examples, e.g., FLAN aggregates numerous open-source datasets and OpenAI leverages examples submitted in the browser playgro…
View article: Draws and windfalls: Comparing patient experiences in inpatient telehealth and non-telehealth acute care units
Draws and windfalls: Comparing patient experiences in inpatient telehealth and non-telehealth acute care units Open
The global COVID-19 pandemic has challenged health care delivery in many ways. One solution from the pandemic that offers potential upside is the expansion of telehealth as a means to provide high quality, cost-effective, and safe care whi…
View article: Validation of a Remote Monitoring Blood Pressure Device in Pregnancy
Validation of a Remote Monitoring Blood Pressure Device in Pregnancy Open
Background The Ideal Life Blood Pressure Manager measures blood pressure (BP) and automatically transmits results to the patient’s medical record independent of internet access, but has not been validated. Our objective was to conduct a va…
View article: Extracting Training Data from Diffusion Models
Extracting Training Data from Diffusion Models Open
Image diffusion models such as DALL-E 2, Imagen, and Stable Diffusion have attracted significant attention due to their ability to generate high-quality synthetic images. In this work, we show that diffusion models memorize individual imag…