#Reasoning

Evaluating Evaluators: Investigating the Performance of LLM Judges in Personalized Writing Style Evaluation

Researchers evaluated the performance of Large Language Model (LLM) judges in personalized writing tasks and found that the quality of judges is weakly correlated with general LLM capabilities measured by MMLU (Pearson’s ρ ≤ 0.2), and their performance is influenced by writing tasks and evaluation criteria (e.g., general helpfulness vs. stylistic personalization). LLM evaluators were more consistent and performed better in pairwise comparisons than in independent scoring responses. The 'strongest' LLM did not reliably replicate human evaluations, possibly due to being overly sensitive to subtle details overlooked by human annotators and weaker LLMs.

ComfyUI achieving double LLM token generation speed

ComfyUI enhances the inference path of the Qwen3.x series models, doubling the generation speed of the same Checkpoint on RTX 5090 without model modifications or quality loss.

Meta Develops Intelligent Delivery System: AI Optimizes Computational Resource Allocation to Enhance Ad Sorting Performance

Meta's Intelligent Delivery system, through AI optimization of computational resource allocation, has boosted ad sorting performance by approximately 10% since 2017. The system includes a Resource Core Model that predicts the marginal utility of computational resources (6B sparse parameters + 20M dense parameters), multi-layer mechanisms for cross-system resource migration, and a framework that uses AI agents to continuously monitor and detect changes in the response system.

German defense technology company Helsing releases Clarity AI platform, providing target preparation suggestions for weapon systems.

German defense technology company Helsing has launched the Clarity AI platform, which generates target suggestions by integrating sensor data, open-source intelligence, etc. The information is processed by hundreds of AI agents and reviewed by humans, emphasizing European control and human supervision.

Genie with Genie Ontology performs excellently in enterprise data proxy benchmark tests

Databricks Genie Ontology's enterprise context layer brings business understanding to ChatGPT, Claude, Microsoft Copilot Cowork, and Grok Bot

Databricks releases Genie Ontology, an enterprise context layer, combining managed definitions with continuous learning of organizational knowledge to provide business understanding capabilities for AI agents. Genie plugins support ChatGPT, Claude, Microsoft Copilot Cowork, and Grok Bot, enabling teams to maintain business context in Databricks and use it across multiple agents.

Databricks Genie Agents now support analyzing more data sources to transform workflows

Databricks Genie Agents can now perform high-quality business analysis, analyze large volumes of documents, and connect enterprise tools to automate actions by supporting unstructured data analysis and new MCP connectors. The new OpenSharing feature allows users to share managed agents with pre-installed data, instructions, and knowledge with partners, suppliers, and customers, accelerating the generation of external insights.

Hugging Face blog: Token Pooling-aware training for multi-vector embedding models.

Hugging Face's blog introduces the Token Pooling perception training method for multi-vector embedding models, which compresses storage by merging similar tokens through HierarchicalTokenPooling.

Anthropic's IPO roadshow video reveals Claude's multi-chip strategy and computational advantages.

Anthropic disclosed in its IPO roadshow video that the Claude model now supports four major chip platforms: AWS Trainium, Google TPUs, NVIDIA GPU, and AMD GPU, a result of its early multi-chip strategy that reduced the deployment time for new platforms from months to three days.

Google SWE uses 48 AI agents to analyze NASA radar data, discovering unreported Mayan pyramids in Mexico within 72 hours

Google SWE uses 48 AI agents to analyze NASA radar data, discovering unreported Mayan pyramids in Mexico within 72 hours; the agents process 1,200 km² of radar images, identifying suspected pyramids, plazas, and reservoir structures with right angles and sharp corners, which have been confirmed by the National Institute of Anthropology and History of Mexico as not being in the registered records.

US data company sells data worth about $500 million to Chinese AI labs, including reinforcement learning environments to help improve model performance in specific fields.

US data company sold about $500 million data to Chinese AI labs, including reinforcement learning environments to help improve model performance in fields such as coding, networking, finance, law, and biology. These environments simulate the world to allow models to train repeatedly and use scoring standards to optimize output. Chinese labs can save over 70% of inference computing costs annually by purchasing the same RL environments as US companies like Anthropic and OpenAI.

Tree-Ising: Uses bitwise Ising editing to fix the inference loop of large language models.

Tree-Ising fixes the inference loop of large language models by solving sparse Ising/QUBO problems on BF16 parameters, improving mathematical benchmark accuracy with fewer weight edits without gradient or LLM-token costs.

MediaTek 'claims' the iFlow intelligent temperature control system chip platform: all models are equipped with Genio 420

MediaTek empowers the launch of ISHO iFlow smart temperature control system and its Pro version, both equipped with the Genio 420 platform, which is based on a 6nm process, offering 7.2 TOPS AI computing power and 16GB LPDDR5X, supporting BCG contactless monitoring function for dual-person independent monitoring, the standard version priced starting from 9999 yuan.

inclusionAI/ZoomGain-4B released: A fine-grained visual supervision model based on global-to-local visual perception gaps

inclusionAI has released the ZoomGain-4B large language model, achieving a score of 61.08 on ZoomGainBench and performing excellently on multiple fine-grained perception and general perception benchmark tests; the model is built with supervision based on the global-to-local visual perception gap mined from global image details, supporting English and Chinese Q&A, and can be downloaded via Hugging Face.

Infrathon: A benchmark for language intelligent agents used in long-term AI infrastructure optimization in the real world.

Infrathon conducted 150 long-term AI infrastructure optimization task evaluations on six models, consuming a total of 10,000 GPU hours, and found that the model rankings changed with the evaluation duration. 135 tasks achieved further optimization in the latter half of the budget, and Claude Opus 5.5 increased the KV cache compression rate from 6.82% to 19.86%, exceeding human design baselines. Analysis shows that task violations, unreliable testing, and premature completion are the main obstacles to continuous optimization.

OpenAI discussed new ways of working for Codex at the AI Engineers Code Summit.

OpenAI introduced new ways of working for Codex at the AI Engineers' Code Summit, including improvements to cloud functions, voice interaction, and agent collaboration. Codex Cloud 2.0 allows users to run multiple tasks in isolated instances, but there are still usability issues with cloud functions. The voice mode, as a control plane, can manage code repositories, and personal agents dots can serve as safety nets, delegation nodes, and ubiquitous presences.

Tether AI Research has launched the 0.8B parameter BotPsy model, used for device-side intelligent assistants and agent workflows

Tether AI Research has launched the 0.8B parameter BotPsy model, built based on Qwen3.5-0.8B, specifically designed for function calls and agent workflows, suitable for on-device deployment. BotPsy outperforms all ≤1.2B parameter models in 6 tool use and agent benchmark tests, including Qwen3.5-2B-Think, achieving the highest scores in 5 tests, with a 2.5x speed increase and a 3x improvement in reasoning efficiency.

Alibaba Qwen Code 1.0 released: Qoder and Qwen Code integrate open source.

Huawei Kirin 9050 series is equipped with a Da Vinci architecture NPU supporting 30B MoE large models on the edge.