【AI最新動向 2026年1月23日】論文5件・GitHub5件

2026年1月23日2026年2月11日

📝 この記事のポイント

🚀 AI技術の最新動向 – 2026年1月23日世界中から収集したAI・機械学習の最新情報をお届けします 📑 目次 💻 注目のGitHubプロジェクト Infatoshi/batmobile fangchenantek/Psychosis-of-AI buyukakyuz/rig RAZZULLIX/fast_topk_batched Colev2/Neural-Networks 📌 関連記事もチェック 📚 最新研究論文 1. Why Can't I Open My Drawer? Mitigating Object-Driven Shortcuts in Zero-Shot Compositional Action Recognition 著者: Geo Ahn, Inwoong Lee, Taeoh Kim We study Compositional Video Understanding (CVU), where models must recognize verbs and objects and compose them to generalize to unseen combinations. We find that existing Zero-Shot Compositional Action Recognition (ZS-CAR) models fail primarily due to an overlooked failure mode: object-driven verb… 論文を読む → 2. PyraTok: Language-Aligned Pyramidal Tokenizer for Video Understanding and Generation 著者: Onkar Susladkar, Tushar Prakash, Adheesh Juvekar Discrete video VAEs underpin modern text-to-video generation and video understanding systems, yet existing tokenizers typically learn visual codebooks at a single scale with limited vocabularies and shallow language supervision, leading to poor cross-modal alignment and zero-shot transfer. We introd… 論文を読む → 3. LLM-in-Sandbox Elicits General Agentic Intelligence 著者: Daixuan Cheng, Shaohan Huang, Yuxian Gu We introduce LLM-in-Sandbox, enabling LLMs to explore within a code sandbox (i.e., a virtual computer), to elicit general intelligence in non-code domains. We first demonstrate that strong LLMs, without additional training, exhibit generalization capabilities to leverage the code sandbox for non-cod… 論文を読む → 💻 注目のGitHubプロジェクト 1. Infatoshi/batmobile High-performance CUDA kernels for equivariant graph neural networks (MACE, NequIP, Allegro). 10-20x faster than e3nn. ⭐ 39 stars | 🔀 6 forks リポジトリを見る → 2. fangchenantek/Psychosis-of-AI 一份来自2026年的AI精神病理学诊断报告 / A Pathological Diagnosis of AI Civilization ⭐ 29 stars | 🔀 0 forks リポジトリを見る → 3. buyukakyuz/rig Distributed LLM inference across machines over WiFi ⭐ 25 stars | 🔀 2 forks リポジトリを見る → 4. RAZZULLIX/fast_topk_batched High-performance batched Top-K selection for CPU inference. Up to 80x faster than PyTorch, optimized for LLM sampling with AVX2 SIMD. ⭐ 15 stars | 🔀 1 forks リポジトリを見る → 5. Colev2/Neural-Networks Assignments on Neural Networks course at CSD AUTH ⭐ 7 stars | 🔀 0 forks リポジトリを見る → 📚 あわせて読みたい【AI最新動向】一歩先の未来へ！私が触れた最先端AIの世界「プロンプトは戦略だ！」と学んで劇的効率UP！私のAI活用術私が実感！AIで物流の悩み解消、時間もコストも浮いた話。

🚀 AI技術の最新動向 – 2026年1月23日

世界中から収集したAI・機械学習の最新情報をお届けします

📑 目次

💻 注目のGitHubプロジェクト
1. Infatoshi/batmobile
2. fangchenantek/Psychosis-of-AI
3. buyukakyuz/rig
4. RAZZULLIX/fast_topk_batched
5. Colev2/Neural-Networks
📌 関連記事もチェック

📚 最新研究論文

1. Why Can't I Open My Drawer? Mitigating Object-Driven Shortcuts in Zero-Shot Compositional Action Recognition

著者: Geo Ahn, Inwoong Lee, Taeoh Kim

We study Compositional Video Understanding (CVU), where models must recognize verbs and objects and compose them to generalize to unseen combinations. We find that existing Zero-Shot Compositional Action Recognition (ZS-CAR) models fail primarily due to an overlooked failure mode: object-driven verb…

論文を読む →

2. PyraTok: Language-Aligned Pyramidal Tokenizer for Video Understanding and Generation

著者: Onkar Susladkar, Tushar Prakash, Adheesh Juvekar

Discrete video VAEs underpin modern text-to-video generation and video understanding systems, yet existing tokenizers typically learn visual codebooks at a single scale with limited vocabularies and shallow language supervision, leading to poor cross-modal alignment and zero-shot transfer. We introd…

論文を読む →

3. LLM-in-Sandbox Elicits General Agentic Intelligence

著者: Daixuan Cheng, Shaohan Huang, Yuxian Gu

We introduce LLM-in-Sandbox, enabling LLMs to explore within a code sandbox (i.e., a virtual computer), to elicit general intelligence in non-code domains. We first demonstrate that strong LLMs, without additional training, exhibit generalization capabilities to leverage the code sandbox for non-cod…

論文を読む →