Large language models (LLMs) are a prominent subset of AI, built on advanced neural network architectures that can generate new data, including text, images, and audio. LLMs utilize various technologies to identify patterns in a given set of training data, without requiring explicit instructions about what to look for [ 12, 35 ] . LLMs typically assume that the training data follows a probability distribution, and once they have identified existing patterns, they can generate new instances that are similar to the original data. By drawing from and combining training data, LLMs can create new content that tran scends the initial dataset [1 7 ].

Federally funded research and development centers (FFRDCs) face text-heavy workloads, from policy documents to scientific and engineering papers, that are slow to analyze manually. We show how large language models can accelerate summarization, classification, extraction, and sense-making with only a few input-output examples. To enable use in sensitive government contexts, we apply OnPrem$.$LLM, an open-source framework for secure and flexible application of generative AI. Case studies on defense policy documents and scientific corpora, including the National Defense Authorization Act (NDAA) and National Science Foundation (NSF) Awards, demonstrate how this approach enhances oversight and strategic analysis while maintaining auditability and data sovereignty.

The generation of high-quality candidate molecules remains a central challenge in AI-driven drug design. Current phenotype-based and target-based strategies each suffer limitations, either incurring high experimental costs or overlook system-level cellular responses. To bridge this gap, we propose ExMoIRL, a novel generative framework that synergistically integrates phenotypic and target-specific cues for de novo molecular generation. The phenotype-guided generator is first pretrained on expansive drug-induced transcriptional profiles and subsequently fine-tuned via multi-objective reinforcement learning (RL). Crucially, the reward function fuses docking affinity and drug-likeness scores, augmented with ranking loss, prior-likelihood regularization, and entropy maximization. The multi-objective RL steers the model toward chemotypes that are simultaneously potent, diverse, and aligned with the specified phenotypic effects. Extensive experiments demonstrate ExMoIRL's superior performance over state-of-the-art phenotype-based and target-based models across multiple well-characterized targets. Our generated molecules exhibit favorable drug-like properties, high target affinity, and inhibitory potency (IC50) against cancer cells. This unified framework showcases the synergistic potential of combining phenotype-guided and target-aware strategies, offering a more effective solution for de novo drug discovery.

Vision-Language Models (VLMs) are foundational to critical applications like autonomous driving, medical diagnosis, and content moderation. While Parameter-Efficient Fine-Tuning (PEFT) methods like LoRA enable their efficient adaptation to specialized tasks, these models remain vulnerable to adversarial attacks that can compromise safety-critical decisions. CLIP, the backbone for numerous downstream VLMs, is a high-value target whose vulnerabilities can cascade across the multimodal AI ecosystem. We propose Dynamic Adversarial Curriculum DAC-LoRA, a novel framework that integrates adversarial training into PEFT. The core principle of our method i.e. an intelligent curriculum of progressively challenging attack, is general and can potentially be applied to any iterative attack method. Guided by the First-Order Stationary Condition (FOSC) and a TRADES-inspired loss, DAC-LoRA achieves substantial improvements in adversarial robustness without significantly compromising clean accuracy. Our work presents an effective, lightweight, and broadly applicable method to demonstrate that the DAC-LoRA framework can be easily integrated into a standard PEFT pipeline to significantly enhance robustness.

The fundamental units of internal representations in large language models (LLMs) remain undefined, limiting further understanding of their mechanisms. Neurons or features are often regarded as such units, yet neurons suffer from polysemy, while features face concerns of unreliable reconstruction and instability. To address this issue, we propose the Atoms Theory, which defines such units as atoms. We introduce the atomic inner product (AIP) to correct representation shifting, formally define atoms, and prove the conditions that atoms satisfy the Restricted Isometry Property (RIP), ensuring stable sparse representations over atom set and linking to compressed sensing. Under stronger conditions, we further establish the uniqueness and exact $\ell_1$ recoverability of the sparse representations, and provide guarantees that single-layer sparse autoencoders (SAEs) with threshold activations can reliably identify the atoms. To validate the Atoms Theory, we train threshold-activated SAEs on Gemma2-2B, Gemma2-9B, and Llama3.1-8B, achieving 99.9% sparse reconstruction across layers on average, and more than 99.8% of atoms satisfy the uniqueness condition, compared to 0.5% for neurons and 68.2% for features, showing that atoms more faithfully capture intrinsic representations of LLMs. Scaling experiments further reveal the link between SAEs size and recovery capacity. Overall, this work systematically introduces and validates Atoms Theory of LLMs, providing a theoretical framework for understanding internal representations and a foundation for mechanistic interpretability. Code available at https://github.com/ChenhuiHu/towards_atoms.

Integrating natural language (NL) prompts into robotic mission planning has attracted significant interest in recent years. In the construction domain, Building Information Models (BIM) encapsulate rich NL descriptions of the environment. We present a novel framework that fuses NL directives with BIM-derived semantic maps via a Beta-Bernoulli Bayesian fusion by interpreting the LLM as a sensor: each obstacle's design-time repulsive coefficient is treated as a Beta(alpha, beta) random variable and LLM-returned danger scores are incorporated as pseudo-counts to update alpha and beta. The resulting posterior mean yields a continuous, context-aware repulsive gain that augments a Euclidean-distance-based potential field for cost heuristics. By adjusting gains based on sentiment and context inferred from user prompts, our method guides robots along safer, more context-aware paths. This provides a numerically stable method that can chain multiple natural commands and prompts from construction workers and foreman to enable planning while giving flexibility to be integrated in any learned or classical AI framework. Simulation results demonstrate that this Beta-Bernoulli fusion yields both qualitative and quantitative improvements in path robustness and validity.

In response to adversarial text attacks, attack detection models have been proposed and shown to successfully identify text modified by adversaries. Attack detection models can be leveraged to provide an additional check for NLP models and give signals for human input. However, the reliability of these models has not yet been thoroughly explored. Thus, we propose and test a novel attack setting and attack, RedHerring. RedHerring aims to make attack detection models unreliable by modifying a text to cause the detection model to predict an attack, while keeping the classifier correct. This creates a tension between the classifier and detector. If a human sees that the detector is giving an ``incorrect'' prediction, but the classifier a correct one, then the human will see the detector as unreliable. We test this novel threat model on 4 datasets against 3 detectors defending 4 classifiers. We find that RedHerring is able to drop detection accuracy between 20 - 71 points, while maintaining (or improving) classifier accuracy. As an initial defense, we propose a simple confidence check which requires no retraining of the classifier or detector and increases detection accuracy greatly. This novel threat model offers new insights into how adversaries may target detection models.

Diplomatic events consistently prompt widespread public discussion and debate. Public sentiment plays a critical role in diplomacy, as a good sentiment provides vital support for policy implementation, helps resolve international issues, and shapes a nation's international image. Traditional methods for gauging public sentiment, such as large-scale surveys or manual content analysis of media, are typically time-consuming, labor-intensive, and lack the capacity for forward-looking analysis. We propose a novel framework that identifies specific modifications for diplomatic event narratives to shift public sentiment from negative to neutral or positive. First, we train a language model to predict public reaction towards diplomatic events. To this end, we construct a dataset comprising descriptions of diplomatic events and their associated public discussions. Second, guided by communication theories and in collaboration with domain experts, we predetermined several textual features for modification, ensuring that any alterations changed the event's narrative framing while preserving its core facts.We develop a counterfactual generation algorithm that employs a large language model to systematically produce modified versions of an original text. The results show that this framework successfully shifted public sentiment to a more favorable state with a 70\% success rate. This framework can therefore serve as a practical tool for diplomats, policymakers, and communication specialists, offering data-driven insights on how to frame diplomatic initiatives or report on events to foster a more desirable public sentiment.

Here's How to Find Out if You're Eligible In a settlement with the FTC, Amazon will have to pay out over a billion dollars to US customers for "deceptive" sign-up and cancellation processes. Amazon customers with a Prime subscription will soon be able to make claims online for their share of the $1.5 billion the company is being ordered to pay to users in the United States. Amazon now has to "provide $1.5 billion in refunds back to consumers harmed by their deceptive Prime enrollment practices," according to a press release from the FTC. The total settlement with the FTC is $2.5 billion, which includes a $1 billion penalty owed to the government. "There was no admission of guilt in this settlement by the company or any executives," says Alisa Carroll, an Amazon spokesperson, in an email sent to WIRED on Thursday after the decision was released.

Amazon Will Pay $2.5 Billion to Settle FTC Suit That Alleged'Dark Patterns' in Prime Sign-Ups Amazon will pay both the Federal Trade Commission and consumers directly to settle a lawsuit alleging that it used manipulative and deceptive tactics to encourage sign-ups for Prime. Amazon has agreed to pay $2.5 billion to settle a lawsuit filed by the Federal Trade Commission, which alleged that the company has "knowingly duped" millions of people into enrolling in its Amazon Prime membership program by using what the FTC has described as " dark patterns, " or, manipulative, coercive, or deceptive user-interface designs." The settlement claimed that Amazon "obtains consumers' billing information before it discloses all material terms for an Amazon Prime subscription," and in doing so, was in violation of the Restore Online Shoppers' Confidence Act, which was signed into law in 2010 to prevent the use of deception to prompt or encourage online purchases. The $2.5 billion payment includes $1 billion that has to be paid to the FTC, and $1.5 billion that will go directly to consumers who unknowingly signed up for Prime, or tried and failed to cancel their Prime subscriptions due to Amazon's online interface, between June 23, 2019 and June 23, 2025. Individual consumers can get compensated up to $51 each. In a statement released by the FTC on Tuesday, agency chairman Andrew Ferguson said that the settlement "made history and secured a record-breaking, monumental win for the millions of Americans who are tired of deceptive subscriptions that feel impossible to cancel." "Today, we are putting billions of dollars back into Americans' pockets, and making sure Amazon never does this again," Ferguson said. Amazon spokesperson Alisa Carroll tells WIRED that there was "no admission of guilt in this settlement by the company or any executives.