As Large Language Models (LLMs) continue to advance, Retrieval-Augmented Generation (RAG) has emerged as a vital technique to enhance factual accuracy by integrating external knowledge into the generation process. However, LLMs often fail to faithfully integrate retrieved evidence into their generated responses, leading to factual inconsistencies. To quantify this gap, we introduce Entity-Context Divergence (ECD), a metric that measures the extent to which retrieved information is accurately reflected in model outputs. We systematically evaluate contemporary LLMs on their ability to preserve factual consistency in retrieval-augmented settings, a capability we define as RAG-ability. Our empirical analysis reveals that RAG-ability remains low across most LLMs, highlighting significant challenges in entity retention and context fidelity. This paper introduces Radiant (Retrieval AugmenteD entIty-context AligNmenT), a novel framework that merges RAG with alignment designed to optimize the interplay between retrieved evidence and generated content. Radiant extends Direct Preference Optimization (DPO) to teach LLMs how to integrate provided additional information into subsequent generations. As a behavior correction mechanism, Radiant boosts RAG performance across varied retrieval scenarios, such as noisy web contexts, knowledge conflicts, and hallucination reduction. This enables more reliable, contextually grounded, and factually coherent content generation.

Language models are prone to occasionally undesirable generations, such as harmful or toxic content, despite their impressive capability to produce texts that appear accurate and coherent. This paper presents a new two-stage approach to detect and mitigate undesirable content generations by rectifying activations. First, we train an ensemble of layerwise classifiers to detect undesirable content using activations by minimizing a smooth surrogate of the risk-aware score. Then, for contents that are detected as undesirable, we propose layerwise distributional intervention policies that perturb the attention heads minimally while guaranteeing probabilistically the effectiveness of the intervention. Benchmarks on several language models and datasets show that our method outperforms baselines in reducing the generation of undesirable output.

Details are in the COPYING file.

Advances in sensor technology, cloud computing, and machine learning (ML) continue to converge to accelerate innovation in the field of remote sensing. However, fundamental tools and technologies still need to be developed to drive further breakthroughs and to ensure that the Global Development Community (GDC) reaps the same benefits that the commercial marketplace is experiencing. This process requires us to take a collaborative approach. Data collaborative innovation -- that is, a group of actors from different data domains working together toward common goals -- might hold the key to finding solutions for some of the global challenges that the world faces. That is why Radiant.Earth is investing in new technologies such as Cloud Optimized GeoTiffs, Spatial Temporal Asset Catalogues (STAC), and ML. Our approach to advance ML for global development begins with creating open libraries of labeled images and algorithms.

Radiant is a robust tool for business analytics and running sophisticated models without any need for code development. It leverages the functions and tools in R and at the same time provides a user-friendly interface. With Radiant, you can manipulate and visualize your data, run different models from simple OLS to decision trees (CART) and neural networks, and evaluate your results. The application is based on the Shiny package and can be run locally or on a server. Radiant was developed by Vicent Nijs.