AI Wiki — Knowledge Base

Sakana AIは、SMBCグループと共同で「提案書自動生成アプリケーション」を開発。三井住友銀行にて実務への適用を開始します。 https://sakana.ai/smbc-proposal-ai/ 複数の「AIエージェント」が自律的に連携し、情報収集から仮説構築、提案構成ま…

Sakana AIは、SMBCグループと共同で「提案書自動生成アプリケーション」を開発。三井住友銀行にて実務への適用を開始します。 https://sakana.ai/smbc-proposal-ai/ 複数の「AIエージェント」が自律的に連携し、情報収集から仮説構築、提案構成までを支援。日本を支える基幹産業のDXを、独自の技術で強力に後押しします。 2025年のパートナーシップ締結以来、検討を

researchdavid-ha--x

30 Apr 2026

All entries

ReGATE: Learning Faster and Better with Fewer Tokens in MLLMs

ReLoop: Structured Modeling and Behavioral Verification for Reliable LLM-Based Optimization

Replit is turning 10, and we are making Agent free Get ready to have fun building! Want prizes? Join our buildathon in partnership with Anth…

🎂 Replit turns 10 this weekend, and we're giving every Replit user free Agent for 24 hours on Saturday.🤯 Clear your Saturday. And tune in …

Report of the 5th PVUW Challenge: Towards More Diverse Modalities in Pixel-Level Understanding

Researchers try to cut the genetic code from 20 to 19 amino acids

resharing this note, find it helpful given all the great open evals work + teams building vertical agents Evals are a proxy for the behavior…

Resume-ing Control: (Mis)Perceptions of Agency Around GenAI Use in Recruiting Workflows

Rethinking KV Cache Eviction via a Unified Information-Theoretic Objective

Rethinking the Harmonic Loss via Non-Euclidean Distance Layers

Retrieval-Augmented LLMs for Evidence Localization in Clinical Trial Recruitment from Longitudinal EHR Narratives

RetroMotion: Retrocausal Motion Forecasting Models are Instructable

Revisiting Human-in-the-Loop Object Retrieval with Pre-Trained Vision Transformers

reward-lens: A Mechanistic Interpretability Library for Reward Models

RFK Jr. appeals ruling that wiped out his vaccine advisory panel

Risk Reporting for Developers' Internal AI Model Use

RL unknotter, hard unknots and unknotting number

Roblox reports Q1 bookings up 43% YoY to 1.7B, vs. 1.73B est., and DAUs up 35% to 132M, below analysts' estimates of 143.8M; RBLX drops 16%+ after hours (Cecilia D'Anastasio/Bloomberg)

robotics announcement next week. who are the best journalists covering the topic?

Robust Alignment: Harmonizing Clean Accuracy and Adversarial Robustness in Adversarial Training

Robust Federated Learning under Adversarial Attacks via Loss-Based Client Clustering

Robust Representation Learning through Explicit Environment Modeling

Roku reports Q1 revenue up 22% YoY to 1.25B, ad revenue up 27% to 613M, subscription revenue up 30%, raises 2026 profit guidance; ROKU jumps 8%+ after hours (Todd Spangler/Variety)

RoseCDL: Robust and Scalable Convolutional Dictionary Learning for Rare event and Anomaly Detection

Rule-based High-Level Coaching for Goal-Conditioned Reinforcement Learning in Search-and-Rescue UAV Missions Under Limited-Simulation Training

Runpod launches Flash to bring AI inference to developers without infra overhead

Safe Navigation using Neural Radiance Fields via Reachable Sets

SafeReview: Defending LLM-based Review Systems Against Adversarial Hidden Prompts

Safety Is Not Universal: The Selective Safety Trap in LLM Alignment

SAGE: A Strategy-Aware Graph-Enhanced Generation Framework For Online Counseling

Sample Selection Using Multi-Task Autoencoders in Federated Learning with Non-IID Data

Samsung says the memory shortage will worsen in 2027 as customers are already placing orders for that year, leading to a projected supply gap larger than 2026 (Kim Jaewon/Nikkei Asia)

SAND: Spatially Adaptive Network Depth for Fast Sampling of Neural Implicit Surfaces

Sandisk reports Q3 revenue up 251% YoY to 5.95B, vs. 4.72B est., consumer revenue below est., and forecasts Q4 revenue above est.; SNDK drops 5%+ after hours (Britney Nguyen/MarketWatch)

Say hello to the Cohere Centre! 🇨🇦 We’re proud to partner with Ottawa’s premier convention and event facility as it enters an exciting new…

SciHorizon-DataEVA: An Agentic System for AI-Readiness Evaluation of Heterogeneous Scientific Data

SciMDR: Advancing Scientific Multimodal Document Reasoning

SD2AIL: Adversarial Imitation Learning from Synthetic Demonstrations via Diffusion Models

SEAL: Semantic-aware Single-image Sticker Personalization with a Large-scale Sticker-tag Dataset

Seamless Indoor-Outdoor Mapping for INGENIOUS First Responders

SecMate: Multi-Agent Adaptive Cybersecurity Troubleshooting with Tri-Context Personalization

Seeking Consensus: Geometric-Semantic On-the-Fly Recalibration for Open-Vocabulary Remote Sensing Semantic Segmentation

Select to Think: Unlocking SLM Potential with Local Sufficiency

Self-Jailbreaking: Language Models Can Reason Themselves Out of Safety Alignment After Benign Reasoning Training

Semantic Embeddings of Chemical Elements for Enhanced Materials Inference and Discovery

Semantic Foam: Unifying Spatial and Semantic Scene Decomposition

Semi-supervised learning with max-margin graph cuts