Research & Papers

Explains the definition of agentic RAG and its differences from traditional pipelines, while examining when vector databases are effective and how to design document features for agents.

• •Explains what agentic RAG is and how it differs from traditional retrieve-then-generate pipelines.
• •Mentions scenarios where vector databases are still meaningful.
• •Proposes how to design document features for AI agents.

Google DeepMind, Meta, and Isomorphic Labs jointly invest $300 million in Biohub, a nonprofit founded by the Zuckerbergs, to support AI dataset construction.

• •Google DeepMind, Meta, and Isomorphic Labs jointly invested $300 million in Biohub, a nonprofit founded by Mark and Priscilla Chan Zuckerberg.
• •This funding is part of a $1.8 billion initiative to build AI datasets that enable researchers to make predictions and ask questions.
• •According to Reuters, the investment aims to accelerate the application of AI in biomedical research.

11個の正方形のパッキングにおける最適性証明のLeanによる形式化が公開され、AIの支援を受けた研究として紹介されている

• •11個の正方形のパッキングにおける最適性証明がLeanで形式化されている
• •この研究はAIの支援を受けた証明として位置づけられている
• •形式化されたコードはQueuingtheorydotcom/11SquaresFormalizedリポジトリで公開されている

Research shows that a single instruction for telegraphic format reduces API output tokens by 40–49% while allowing full information restoration across four model families.

• •Specifying telegraphic format, which excludes articles and redundant expressions, via a single instruction reduces API output tokens by 40–49%.
• •It was confirmed that information can be restored with full fidelity from the compressed format across four different model families.
• •The potential to halve output costs for major APIs in machine-to-machine communication was demonstrated.

OpenWAM, a shared framework integrating video-action programs and local context dynamics, is evaluated through robot control, frozen component transfer, and action-conditioned prediction.

• •It is designed as a shared framework handling video-action programs and local context dynamics.
• •Evaluation was conducted from three perspectives: robot control, frozen component transfer, and action-conditioned prediction.
• •It is an open framework aimed at realizing composable world-action models.

Researchers from Microsoft and the National University of Singapore published a paper investigating the vulnerabilities and causes in applications automatically generated by AI.

• •Researchers affiliated with Microsoft and the National University of Singapore published a paper.
• •The paper is titled "Understanding the (In)Security of Vibe-Coded Applications."
• •It investigates the reality and causes of vulnerabilities in AI-auto-generated applications.

OpenAI revealed that it has published 722 mathematical results generated by its frontier models on GitHub, many of which include formal proofs using the Lean proof assistant.

• •722 mathematical results generated by internal frontier models have been published on GitHub.
• •Many of the results include formal proofs using the Lean proof assistant.
• •Referencing recommendations from the independent organization AGMAI, summaries of computational complexity and reasoning processes are also disclosed.

News that OpenAI has released 722 manuscripts related to mathematics.

OpenAIが未公開のフロンティアモデルによって解決された長年の数学的課題を含む722本の原稿を公開し、数学コミュニティの関心と倫理的問題を提起している

• •未公開のフロンティアモデルによって生成された722本の原稿が公開された
• •原稿は関連する論文をグループ化した372の結果ファミリーをカバーしている
• •数学コミュニティの一部からは感銘を受けたという声と、研究倫理に関する懸念が同時に上がっている

The article explains the proposal of UniEvo-VL, a self-evolving framework where multimodal models learn from self-feedback, and the use of self-correction in test-time computation.

• •It leverages the ability of integrated multimodal systems for generation and understanding to learn from their own feedback.
• •It utilizes the model's own self-criticism as privileged information without relying on external teacher models.
• •It improves the model through constructive self-correction feedback during the test-time computation stage.