The future of AI and jobs will be so much weirder than you think. This article was featured in the One Story to Read Today newsletter. Last year, Steve Yegge started "suddenly getting pounded by nap attacks in the middle of the day." Without fail, Yegge--a programmer and tech blogger--would "hit a wall, fall over, and sleep for 90 minutes," he told me. Like many developers, Yegge no longer writes code by hand; instead, he manages a legion of bots to do that for him.

Ukraine launched one of its largest drone attacks on Moscow, striking a major oil refinery and sending thick black smoke over the capital, according to Russian officials.

Automated interpretability research aims to identify concepts encoded in neural network features to enhance human understanding of model behavior. Within the context of large language models (LLMs) for natural language processing (NLP), current automated neuron-level feature description methods face two key challenges: limited robustness and the assumption that each neuron encodes a single concept (monosemanticity), despite increasing evidence of polysemanticity. This assumption restricts the expressiveness of feature descriptions and limits their ability to capture the full range of behaviors encoded in model internals. To address this, we introduce Polysemantic FeatuRe Identification and Scoring Method (PRISM), a novel framework specifically designed to capture the complexity of features in LLMs. Unlike approaches that assign a single description per neuron, common in many automated interpretability methods in NLP, PRISM produces more nuanced descriptions that account for both monosemantic and polysemantic behavior. We apply PRISM to LLMs and, through extensive benchmarking against existing methods, demonstrate that our approach produces more accurate and faithful feature descriptions, improving both overall description quality (via a description score) and the ability to capture distinct concepts when polysemanticity is present (via a polysemanticity score).

The software engineers filed a complaint with Seattle's civil rights office accusing Amazon of illegally retaliating against them for expressing their personal political beliefs. Earlier this month, five current Amazon employees publicly urged Seattle City Council to regulate data centers . It was an unprecedented act of advocacy by tech workers, and now three of the staffers say they are under internal investigation for what they understand to be allegedly representing themselves as spokespeople for the company without prior approval. "It's a totally ridiculous claim," says one of the affected employees, Patrick Schloesser. The three software engineers, who work in different divisions of Amazon and all live in Seattle, believe they are being unfairly targeted for expressing their political beliefs.

Modern language model (LM) training has been divided into multiple stages, making it difficult for downstream developers to evaluate the impact of design choices made at each stage. We present EvoLM, a model suite that enables systematic and transparent analysis of LMs' training dynamics across pre-training, continued pre-training, supervised fine-tuning, and reinforcement learning. We train over 100 LMs with 1B and 4B parameters from scratch, and evaluate both upstream (language modeling) and downstream (problem-solving) capabilities, including considerations of both in-domain and out-of-domain generalization. Key insights highlight the diminishing returns from excessive pre-training and post-training, the importance and practices of mitigating forgetting during domain-specific continued pre-training, the crucial role of continued pre-training in bridging pre-training and post-training phases, and various intricate trade-offs when configuring supervised fine-tuning and reinforcement learning. To facilitate open research and reproducibility, we release all pre-trained and post-trained models, training datasets for all stages, and our entire training and evaluation pipeline.

Test-time compute has empowered multimodal large language models to generate extended reasoning chains, yielding strong performance on tasks such as multimodal math reasoning. However, we observe that this improved reasoning ability often comes with increased hallucination: as generations become longer, models tend to drift away from image-grounded content and rely more on language priors. Attention analysis reveals that longer reasoning chains reduce focus on visual inputs, contributing to hallucination. To systematically study this phenomenon, we introduce RH-AUC, a metric that quantifies how a model's perception accuracy changes with reasoning length, enabling evaluation of whether the model preserves visual grounding while reasoning. We also release RH-Bench, a diagnostic benchmark covering diverse multimodal tasks, designed to jointly assess the balance of reasoning ability and hallucination. We find that (i) larger models generally exhibit a better balance between reasoning and perception; (ii) reasoning and perception balance depends more on the types and domains of the training data than its volume. Our findings highlight the need for evaluation frameworks that account for both reasoning quality and perceptual reliability.

When you purchase through links in our articles, we may earn a small commission. Amazon's Prime Day is one of the best opportunities for shoppers to score great deals on Thunderbolt docks and their cousins, USB-C hubs. I should know -- I've been tracking them for years now. Docking stations and hubs offer an opportunity to connect more peripherals to your desk without breaking the bank. Amazon has been the traditional home of the best deals in both categories, year round, which makes it an even more ideal place to shop during Prime Day -- and before.

Computer-Aided Design (CAD) plays a foundational role in modern manufacturing and product development, often requiring designers to modify or build upon existing models. Converting 3D scans into parametric CAD representations--a process known as CAD reverse engineering--remains a significant challenge due to the high precision and structural complexity of CAD models. Existing deep learning-based approaches typically fall into two categories: bottom-up, geometry-driven methods, which often fail to produce fully parametric outputs, and top-down strategies, which tend to overlook fine-grained geometric details.

Scientific sketches (e.g., models) offer a powerful lens into students' conceptual understanding, yet AI-powered automated assessment of such free-form, visually diverse artifacts remains a critical challenge. Existing solutions often treat sketch evaluation as either an image classification task or monolithic vision-language models, which lack interpretability, pedagogical alignment, and adaptability across cognitive levels. To address these limitations, we present SKETCHMIND, a cognitively grounded, multi-agent framework for evaluating and improving studentdrawn scientific sketches. SKETCHMIND introduces Sketch Reasoning Graphs (SRGs), semantic graph representations that embed domain concepts and Bloom's taxonomy-based cognitive labels. The system comprises modular agents responsible for rubric parsing, sketch perception, cognitive alignment, and iterative feedback with sketch modification, enabling personalized and transparent evaluation. We evaluate SKETCHMIND on a curated dataset of 3,575 student-generated sketches across six science assessment items with different highest order of Bloom's level that require students to draw models to explain phenomena. Compared to baseline GPT-4o performance without SRG(average accuracy: 55.6%), and with bSRGintegration achieves 77.1% average accuracy (+21.4% average absolute gain).

Vision-language-action (VLA) reasoning tasks require agents to interpret multimodal instructions, perform long-horizon planning, and act adaptively in dynamic environments. Existing approaches typically train VLA models in an end-to-end fashion, directly mapping inputs to actions without explicit reasoning, which hinders their ability to plan over multiple steps or adapt to complex task variations. In this paper, we propose ThinkAct, a dual-system framework that bridges high-level reasoning with low-level action execution via reinforced visual latent planning. ThinkAct trains a multimodal LLM to generate embodied reasoning plans guided by reinforcing action-aligned visual rewards based on goal completion and trajectory consistency. These reasoning plans are compressed into a visual plan latent that conditions a downstream action model for robust action execution on target environments. Extensive experiments on embodied reasoning and robot manipulation benchmarks demonstrate that ThinkAct enables few-shot adaptation, long-horizon planning, and self-correction behaviors in complex embodied AI tasks.