xAI's Grok Build CLI was released in early beta, featuring 8 parallel subagents and a 2-million-token context window. Cursor's Composer 2.5 reached general availability, demonstrating parity with advanced models on coding tasks, and Alibaba's Qwen 3.7-Max-Preview introduced a 1-million-token context window with an extended-thinking mode.
RSAT is a new method that trains small language models (SLMs) to produce faithful, step-by-step reasoning for table questions, grounded with cell-level citations. It significantly improves faithfulness (3.7x) and achieves near-perfect citation validity by integrating attribution into the reasoning process.
This paper explores why language models exhibit less surprisal than humans when processing syntactically ambiguous sentences. It tests the hypothesis that LMs can simultaneously consider a greater number of sentence interpretations using Recurrent Neural Network Grammars.
This research paper addresses the limitation of users interacting with language models via single outputs, which hides the full distribution of possible generations. It introduces GROVE, an interactive visualization that represents multiple LM generations as overlapping paths in a text graph, revealing shared structure and branching points for better understanding.
This research explores the intrinsic differences in text generated by Diffusion Language Models (DLMs) and Autoregressive Language Models (ARMs), finding that DLMs show lower n-gram entropy but higher semantic coherence and diversity. Controlled experiments reveal that DLM training objectives enhance coherence and diversity through bidirectional context, while decoding algorithms are responsible for entropy reduction.
This article explores Direct Preference Optimization (DPO), a method for aligning AI models with human preferences, examining its potential applications beyond traditional chatbots. It delves into how DPO can be utilized in various AI domains.
NVIDIA's Nemotron-Labs Diffusion aims to accelerate AI applications by tackling the one-token bottleneck through parallel generation of multiple tokens. This new diffusion language model claims up to 6.4 times higher tokens per forward pass, significantly benefiting latency-sensitive AI products like coding assistants and agent workflows.
This content discusses the development of Nemotron-Labs' diffusion language models aimed at achieving exceptionally high-speed text generation. The focus is on technical advancements to optimize the rapidity of text production.
This research explores the entropy mechanism within reinforcement learning, specifically its application to enhance reasoning capabilities in language models. It investigates how entropy can be leveraged to improve the learning process and decision-making for more robust language model reasoning.
The text discusses the challenge of building language models for critically endangered heritage languages under extreme data sparsity scenarios. The author recounts their personal experience with a minuscule dataset for a language like Halkomelem, highlighting the need for novel approaches for such situations.
The article details OpenHuman's aggressive context-harvesting engine, TokenJuice, which acts as a memory parasite for language models. It continuously extracts, condenses, and reuses context from user activity for future AI inference.
A new Glean benchmark in Claude Cowork indicates that off-the-shelf MCP servers fail 2.5 times more often and use 30% more tokens than Glean's indexed context layer. Users have also reported cutting Claude token bills by 30% by adopting Glean's method.
This research explores molecular representation learning by leveraging language models and incorporating domain-relevant auxiliary tasks. The aim is to enhance the understanding and predictive capabilities for molecular properties.
This article highlights how most English-first Health AI systems fail to understand and respond accurately to medical queries in local languages like Marathi. It emphasizes the need for AI that reasons natively in multiple languages, rather than relying on translation layers or thin localized veneers, to provide effective healthcare guidance.
The author developed Gemini Nano Book Lab, a Chrome extension utilizing the Prompt API to interact with an on-device language model using a lengthy memoir. The experiment investigates the model's behavior, retrieval costs, and API limitations when processing substantial text volumes.
Este artigo explica como os detectores de texto de IA funcionam, analisando a previsibilidade do texto gerado por modelos de linguagem. Ele detalha que textos humanos são estatisticamente mais imprevisíveis e que os detectores exploram essa diferença usando sinais como perplexidade e 'burstiness'.
A new automated pipeline, SkillX, improves the performance of autonomous language-model agents by extracting reusable, hierarchical behaviors from collective trajectories. This three-tiered knowledge base (strategic, functional, atomic skills) allows weaker models to efficiently retrieve experiences, overcoming limitations of traditional methods.
A new study introduces Micro Language Models (μLMs), ultra-compact models (8M–30M parameters) that can deliver large-model quality on devices. This approach solves the dilemma between responsive first words and complete answers in edge assistants by seeding answers locally, reducing latency caused by cloud round-trips.
This content outlines a layered AI workflow to automate complex customs documentation and HS code generation, eliminating manual administrative burdens and improving scalability in international trade. It leverages advanced language models for data extraction and workflow automation platforms for document generation.
CAMO é uma nova técnica de ensemble otimizada para dados desbalanceados, que impulsiona classes minoritárias e melhora o desempenho geral. Avaliada em benchmarks com diferentes modelos de linguagem, CAMO consistentemente atinge a maior pontuação F1 macro, estabelecendo um novo padrão.