large language models

262 items

RESEARCHarXiv CS.CL·4/8/2026

MegaTrain: Full Precision Training of 100B+ Parameter Large Language Models on a Single GPU

MegaTrain é um sistema focado em memória que permite o treinamento eficiente de modelos de linguagem grandes com mais de 100 bilhões de parâmetros em precisão total em uma única GPU. Ele armazena parâmetros na memória do host e utiliza otimizações como um motor de execução pipeline e templates de camada sem estado para superar gargalos de largura de banda e maximizar a utilização da GPU.

Single GPU Training Memory Optimization GPU Acceleration large language models

RESEARCHDEV.to AI·18d ago

Hugging Face: New Research Highlights Value of Specialized AI Models

Hugging Face published research by Dharma AI on May 22, 2026, highlighting that specialized AI models can outperform larger, general-purpose models in specific tasks. The study suggests a strategic shift in AI procurement, emphasizing task-specific performance and efficiency.

specialized AI models Hugging Face AI procurement large language models

RESEARCHarXiv CS.LG·20d ago

ReCrit: Transition-Aware Reinforcement Learning for Scientific Critic Reasoning

ReCrit is a new reinforcement learning framework designed to improve large language models' performance in scientific critic interaction. It addresses the issue of LLMs abandoning correct solutions after user criticism by focusing on inter-turn correctness transitions and categorizing behaviors like correction, sycophancy, and robustness.

reinforcement learning learning Scientific Reasoning large language models

RESEARCHarXiv CS.CL·15d ago

Can AI Guess What You Know? Performance Comparison of Large Language Models for Human Domain Knowledge Estimation From Communication Logs

This research investigates the ability of Large Language Models (LLMs) to infer individual domain knowledge directly from long-term Slack communication logs. Evaluating seven models against self-reported skill ratings, Gemini 2.5 Flash achieved the lowest error, demonstrating the feasibility and current limits of automated expertise mapping.

future-of-work expertise mapping domain knowledge organizational productivity

RESEARCHarXiv CS.CL·5d ago

Computational conceptual history of scientific concepts: From early digital methods to LLMs

This article positions Large Language Models (LLMs) within the history of computational approaches to concept analysis in the history, philosophy, and sociology of science. It examines LLMs' contributions, inherited problems, and reviews recent case studies.

computational conceptual history digital methods concept analysis history of science

RESEARCHarXiv CS.LG·4/23/2026

Expert Upcycling: Shifting the Compute-Efficient Frontier of Mixture-of-Experts

Expert Upcycling proposes a method to progressively expand Mixture-of-Experts (MoE) capacity in large language models during continued pre-training. It increases the number of experts via duplication and router extension to provide a warm initialization, aiming to reduce training costs while preserving per-token inference cost.

Model Architecture training-optimization large language models

ARTICLEDEV.to AI·4/10/2026

AI21 Labs — Deep Dive

AI21 Labs é uma empresa israelense de IA e produto, um player significativo no espaço de IA generativa, competindo com gigantes como OpenAI. A empresa escalou seus modelos de linguagem de 1.5 bilhões para até 398 bilhões de parâmetros, oferecendo produtos como o assistente de escrita Wordtune e o modelo de contexto longo Jamba.

NLP AI products large language models AI21 Labs

ARTICLEDEV.to AI·5/2/2026

From prompt engineering to context engineering

The article proposes a crucial shift from prompt engineering to context engineering, arguing that many AI failures stem from missing relevant information rather than poor phrasing. Context engineering involves deliberately providing the AI with crucial data, such as system instructions, project documentation, and source files, before it acts.

prompt-engineering Context Engineering large language models AI development

ARTICLEDEV.to AI·4/22/2026

A Looming Crisis of AI Generated Text

The article discusses the shift from AI assistance to replacement in text generation, exemplified by new models like Mythos, and its profound impact on literacy and education. The author, who straddles machine learning and literature, rejects the impulse to abandon human writing despite AI's effectiveness.

ethics education future-of-work large language models

ARTICLEDEV.to AI·27d ago

VLAs are dead, long live World Action Models - a summary of Jim Fan's Robotics End Game talk

Jim Fan of Nvidia's robotics group proposes that robotics is entering its "end game" and will follow the same four-stage trajectory as large language models. He asserts that "robotics is entering its end game, and the playbook is already written" by LLMs.

future-of-AI AI large language models robotics

RESEARCHarXiv CS.AI·4/14/2026

Help Without Being Asked: A Deployed Proactive Agent System for On-Call Support with Continuous Self-Improvement

This paper introduces Vigil, a novel proactive AI agent system designed to support human analysts during on-call interactions in large-scale cloud service platforms. Unlike reactive agents, Vigil remains engaged throughout the entire resolution life-cycle, learning from unresolved cases and providing continuous assistance to reduce human workload.

On-Call Support proactive AI customer support large language models

RESEARCHarXiv CS.CL·4/14/2026

Generating High Quality Synthetic Data for Dutch Medical Conversations

This paper presents a pipeline for generating synthetic Dutch medical dialogues using a fine-tuned Large Language Model to address the scarcity of clinical data due to privacy constraints. Evaluations showed strong lexical variety but a scripted conversation flow and issues in domain specificity during qualitative review.

synthetic data Clinical Communication Dutch Language Medical NLP

RESEARCHarXiv CS.AI·4d ago

What Should Agents Say? Action-state Communication for Efficient Multi-Agent Systems

This paper analyzes inter-agent communication strategies in multi-agent systems built on large language models, finding that unconstrained natural language can inflate token usage and affect performance. It proposes PACT (Protocolized Action-state Communication and Transmission), a method to optimize communication by projecting raw agent outputs into compact action-state records.

Communication protocols efficiency Token usage multi-agent systems

NEWSDEV.to AI·4/18/2026

Large Language Letters 04/18/2026

Anthropic's Claude Opus 4.7 showed significant advancements across various benchmarks such as SWEBench Pro, GDP Val, and vision capabilities. The model surpassed previous versions and competitors in several metrics, though independent observers noted some regressions.

AI models Benchmarking Anthropic large language models

ARTICLEDEV.to AI·4/23/2026

how to run qwen3.6-27b locally — the dense 27B that beats the 35B MoE on coding

Alibaba has released Qwen3.6-27B, a 27-billion parameter dense model that outperforms its previous MoE version on coding tasks. This content details how to run the model locally using Ollama, including commands for various quantizations and hardware requirements.

Ollama Local AI model deployment large language models

RESEARCHarXiv CS.AI·4/7/2026

Evaluating Artificial Intelligence Through a Christian Understanding of Human Flourishing

Este conteúdo argumenta que o alinhamento de IA é um problema de formação, não apenas de segurança, pois LLMs atuam como instrumentos de catequese digital que moldam o entendimento humano. É introduzido o Flourishing AI Benchmark (FAI-C-ST) para avaliar modelos de IA contra uma compreensão cristã do florescimento humano, revelando que os sistemas atuais não são neutros, mas aderem a um Secularismo Processual.

AI alignment Avaliação de Modelos Filosofia da IA Ética em IA

RESEARCHarXiv CS.AI·4/7/2026

Toward Full Autonomous Laboratory Instrumentation Control with Large Language Models

Este trabalho explora o potencial de Grandes Modelos de Linguagem (LLMs), como o ChatGPT, e agentes de IA para automação e controle de instrumentação laboratorial. Demonstra-se como essas ferramentas reduzem barreiras de programação e podem evoluir para agentes autônomos capazes de operar equipamentos científicos e refinar estratégias de controle.

LLMs ChatGPT Instrumentation Control large language models

ARTICLEDEV.to AI·21d ago

Mastering the Art of Conversation: Expert ChatGPT Tips and Tricks

This article explores expert tips and tricks for mastering ChatGPT, OpenAI's revolutionary AI chatbot. It discusses understanding its capabilities and limitations to unlock its full potential in conversations and various applications.

learning ChatGPT NLP AI

ARTICLEDEV.to AI·5d ago

MiniMax M3: An Open-Weight Frontier Model You Can Self-Host

The MiniMax M3 is introduced as the first open-weight frontier model combining advanced coding, a 1M-token context window, and native multimodality. It leads the open-weight SWE-Bench Pro leaderboard, offering benefits like no per-token API charges and data residency for self-hosting.

multimodal AI self-hosting Open-weight AI AI benchmarking

RESEARCHarXiv CS.AI·4/20/2026

LLM Reasoning Is Latent, Not the Chain of Thought

This position paper argues that large language model (LLM) reasoning should be studied as latent-state trajectory formation rather than faithful surface chain-of-thought (CoT). It formalizes three competing hypotheses regarding the primary object of reasoning, impacting claims about faithfulness, interpretability, and benchmarks.

Chain-of-Thought interpretability AI Reasoning large language models