StateBench evaluates how language models track state changes across lengthy documents
StateBench tests long-context LLM state tracking across four difficulty levels, revealing where architectures fail and aiding model selection.
StateBench tests long-context LLM state tracking across four difficulty levels, revealing where architectures fail and aiding model selection.
LLM 0.37 is a consolidation release for the CLI and Python library. It adds gpt-6.1-sol and gpt-6.1-terra, changes the default model to the cheaper gpt-6-luna, and introduces llm openai endpoint add f…
A preprint shows compressed shared memory in LLM groups can keep misleading decoys, overweight apparent consensus, and fail across four models.
Allen Institute benchmark tests LLM novelty judges using top conference peer reviews. Prompt wording caused accuracy swings, limiting reliability.
Constrained decoding affects LLM outputs differently by type; drops reflect prompting/format mismatches, not masking, while some constraints help.
OpenAI LLM Terra offers mid-tier pricing, visual reasoning, computer use, and natural writing for mixed tasks, while cheaper automation favors Luna.
Gemini 4 Argon targets cybersecurity, coding, enterprise work with $2/$10 per million input/output, 1 million token limit, and top benchmarks.
LLM post-training shapes behavior through supervised fine-tuning, preference data, and verifiable rewards, making judging easier than generating.
NESTONE trains one LLM with early exits and shared layers, letting decisions stop at needed depth and cut compute while keeping accuracy.
WhiskeyJack tested 240 LLM answers and found six traps in brand mention measurement: inflated citations, prompt bias, noise, and ranking errors.
Four context layers isolate agents, limit prompts, use files, and verify outputs to keep production AI agents within 200,000 tokens.
OpenAI launched GPT-6 Sol and GPT-6 Luna, two API-only models positioned below flagship GPT-6 Astra. Sol handles complex coding and professional tasks; Luna powers fast, high-volume work. Both replace…
A beginner guide defines AI agents as LLMs using tools in loops, compares chatbots, explains ReAct, and warns about risky actions like Replit.
Meta 发布 Muse Spark Open,一个 816B 参数的 MoE 模型,拥有 42B 活跃参数和 1M token 上下文,专为代理编程工作流设计。它在 NVIDIA GB300 NVL72 上训练和优化,支持多步任务执行、工具调用、子代理协调和原生多模态感知。稀疏 MoE 路由每次激活 42B 参数,在 Blackwell Ultra 上实现每 GPU 50K tokens/s 和…
Low-Rank Adaptation updates frozen LLM layers and vision reads screenshots, improving department accuracy, calibration, and memory handling.
A diagnostic quiz for Python LLM apps, checking APIs, prompts, LangChain, RAG, embeddings, vector databases, agents, and MCP tool connections.
GPT-5 is now available on the API platform, bringing significantly stronger reasoning, improved performance on real-world coding tasks like debugging and code generation, and new developer controls fo…
A four-part LangChain guide shows how to diagnose and fix RAG retrieval failures in chunking, embeddings, query handling, ranking, and data quality.
RSI suggests LLMs may help improve their own versions, shifting AI development from human-only engineering toward assisted, recursive iteration.
OpenAI has introduced GPT-6 Astra, described as its most intelligent and best-aligned model to date. The flagship model leads in four key areas: computer use, coding, cybersecurity, and science. Beyon…