[NeurIPS 2025] Thinkless: LLM Learns When to Think

[ICLR 2026]🌴 ARES is an open-source framework for adaptive multimodal reasoning, featuring a two-stage pipeline—Adaptive Cold-Start and Entropy-Shaped Policy Optimization—to balance reasoning depth and efficiency.

A powerful multi-agent reasoning system with temperature-driven cognitive styles. Control search behavior from rigorous academic analysis to creative exploration through simple configuration.

This engine models adaptive reasoning by integrating metacognitive feedback, enabling systems to refine their decision-making through self-evaluation and dynamic restructuring. 本エンジンはメタ認知的フィードバックを統合し、自己評価と動的再構成を通じて意思決定を洗練させる適応的推論をモデル化します。

A constructive framework that systematically integrates causal structures across domains, enabling adaptive reasoning, predictive modeling, and cross-contextual understanding in AI and data systems. 構成的因果統合理論は、因果構造を領域横断的に統合し、AIやデータシステムにおける適応的推論・予測モデリング・文脈横断的理解を可能にする枠組みです。

This theory defines a mechanism by which agents recursively align their current and anticipated intentions through hierarchical feedback and contextual reasoning. It supports robust goal consistency in multi-agent and adaptive systems. 本理論は、エージェントが現在および予測される意図を階層的フィードバックと文脈的推論を通じて再帰的に整合させる仕組みを定義します。マルチエージェントおよび適応システムにおいて強固な目標整合性を支援します。

This model defines a hierarchical framework where emotional signals are processed in multiple stages to influence cognitive behavior. It enables adaptive reasoning by integrating affective feedback into decision-making and memory modulation. 本モデルは、感情信号を多段階で処理し、認知行動に影響を与える階層的枠組みを定義します。感情的フィードバックを意思決定や記憶調整に統合することで、適応的な推論を可能にします。