|
Sakana AIとUC Berkeleyの研究チームが2026年10月12日に共同発表したMASS(Multi-Agent Self-Supervision:マルチエージェント自己監督)は、単一のAIモデルが自律的に複数インスタンスによるチームを編成し、その協調作業の履歴を通じて自己学習を反復する革新的な再帰的自己改善アーキテクチャです。モデル自身がタスク解決者、評価者、組織改善者の役割を兼任しながら、動的なワークフローの進化とパラメータへの知識蒸留を自律的に遂行します。この手法により、マルチエージェント協調の軌跡は単一エージェントの長大な推論よりも高い学習効率を発揮し、オープンエンドな研究タスク群で大幅な性能向上を実証したということです。 生成AIに、MASS(Multi-Agent Self-Supervision:マルチエージェント自己監督)を深掘りさせ、この分野での日米比較も検討させましたので、ご参照ください。 なお、生成AIによる調査・分析結果は、公開された情報だけに基づくものであり、必ずしも実情を示したものではなく、誤った情報を含む可能性があることにご留意のうえ、ご参照ください。 Recursive Self-Improvement through Collective Intelligence October 12, 2026 https://sakana.ai/mass/ Sakana AI’s “MASS” Technology MASS (Multi-Agent Self-Supervision), jointly announced by research teams at Sakana AI and UC Berkeley on October 12, 2026, is an innovative architecture for recursive self-improvement in which a single AI model autonomously assembles a team comprising multiple instances of itself and iteratively learns from the records of their collaborative work. The model itself serves as task solver, evaluator, and optimizer of team organization, autonomously evolving its workflows and distilling knowledge into its own parameters. The researchers report that, with this approach, multi-agent collaboration trajectories enable more efficient learning than lengthy reasoning traces produced by a single agent, and that the method has demonstrated substantial performance improvements across a range of open-ended research tasks. I have asked generative AI to conduct an in-depth analysis of MASS (Multi-Agent Self-Supervision) and to examine how Japan and the United States compare in this field. The resulting analysis is provided for your reference. Please note that the research and analysis produced by generative AI are based solely on publicly available information, do not necessarily reflect the actual state of affairs, and may contain inaccuracies. Please bear this in mind when reviewing the material. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document.
0 Comments
Leave a Reply. |
著者萬秀憲 アーカイブ
April 2026
カテゴリー |
RSS Feed