U
Uri Maoz
How robots learn: A brief, contemporary history
3dUri Maoz
Filtered by #research-breakthroughClear
R
Roni Itkin, Noam Issachar, Yehonatan Keypur, Yehonatan Keypur, Anpei Chen, Sagie BenaimRAD-2: Scaling Reinforcement Learning in a Generator-Discriminator Framework
RAD-2 combines diffusion-based trajectory generation with RL-optimized reranking for autonomous driving motion planning. The framework achieves 56% collision rate reduction through temporal consistency in policy optimization and structured feedback mechanisms. Real-world deployment validates improved safety and smoothness in urban traffic scenarios.See more
4dRoni Itkin, Noam Issachar, Yehonatan Keypur, Yehonatan Keypur, Anpei Chen, Sagie Benaim
R
Roni Itkin, Noam Issachar, Yehonatan Keypur, Yehonatan Keypur, Anpei Chen, Sagie BenaimDR^{3}-Eval: Towards Realistic and Reproducible Deep Research Evaluation
DR^3-Eval is a new benchmark for evaluating deep research agents on complex multi-step tasks using a static corpus that simulates web complexity while remaining reproducible. It introduces a five-dimensional evaluation framework (Information Recall, Factual Accuracy, Citation Coverage, Instruction Following, Depth Quality) and reveals critical failure modes in retrieval robustness and hallucination control.See more
4dRoni Itkin, Noam Issachar, Yehonatan Keypur, Yehonatan Keypur, Anpei Chen, Sagie Benaim
R
Roni Itkin, Noam Issachar, Yehonatan Keypur, Yehonatan Keypur, Anpei Chen, Sagie BenaimGlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens
GlobalSplat introduces a novel 3D Gaussian Splatting framework that reconstructs scenes using global scene tokens instead of pixel-aligned primitives, achieving competitive rendering with only 16K Gaussians (4MB footprint) and 78ms inference. The approach avoids representation bloat via coarse-to-fine training and global scene awareness. Benchmarked on RealEstate10K and ACID, it is significantly faster and more memory-efficient than dense baselines.See more
4dRoni Itkin, Noam Issachar, Yehonatan Keypur, Yehonatan Keypur, Anpei Chen, Sagie Benaim
H
Hugging FaceBuilding a Fast Multilingual OCR Model with Synthetic Data
Hugging Face demonstrates building a fast multilingual OCR model using synthetic data generation. The approach combines data synthesis techniques with model optimization to achieve efficient performance across multiple languages. This method reduces reliance on expensive labeled datasets while maintaining accuracy.See more
3dHugging Face
C
Computational IntelligenceWeekly Review 17 April 2026
Weekly AI roundup curating 22 stories spanning business integration, regulation, and technical failures. Key themes: enterprises spending heavily on AI despite marginal returns (30% of projects pay off), regulators (China, California) tightening safety rules, and persistent reliability concerns (Google AI Overviews wrong 10% of the time, medical AI hallucinates without image input). Widening gap between AI hype and practical business value.See more
3dComputational Intelligence
M
MITCSAIL - YouTubeThe Microscope That Learns What to Look At
MIT CSAIL research on AI-enhanced microscopy systems that learn to identify and focus on scientifically relevant regions autonomously. Demonstrates adaptive machine learning applied to laboratory automation.See more
4dMITCSAIL - YouTube
T
The Verge
Has Google's AI watermarking system been reverse-engineered?
6dThe Verge
U
Uri Maoz
Want to understand the current state of AI? Check out these charts.
7dUri Maoz
H
Hugging FaceInside VAKRA: Reasoning, Tool Use, and Failure Modes of Agents
Hugging Face publishes technical analysis of VAKRA, an agent framework examining reasoning patterns, tool-use capabilities, and failure modes. The post provides insights into how agents handle complex tasks and where they break down.See more
5dHugging Face
U
Uri MaozComing soon: 10 Things That Matter in AI Right Now
MIT Tech Review announces its 2026 Breakthrough Technologies list, covering AI, energy, biotech, and other sectors expected to reshape work and society. This year's selection faced unique challenges due to AI's rapid evolution and broad applicability. Full predictions and analysis coming soon.See more
6dUri Maoz
B
Bala Priya C5 Techniques for Efficient Long-Context RAG
Machine Learning Mastery outlines five optimization techniques for long-context retrieval-augmented generation (RAG) systems. The post covers practical methods to improve efficiency when processing extended document contexts with LLMs. Specific techniques and implementation details are not fully visible in the provided excerpt.See more
5dBala Priya C
U
Uri MaozMustafa Suleyman: AI development won't hit a wall anytime soon—here's why
Mustafa Suleyman argues that AI development will continue accelerating because humans systematically underestimate exponential growth. The piece challenges the assumption that AI progress must plateau, rooted in our evolutionary bias toward linear thinking.See more
12dUri Maoz
S
Sayash Kapoor
Open-world evaluations for measuring frontier AI capabilities
4dSayash Kapoor
U
Uri MaozThe one piece of data that could actually shed light on your job and AI
Silicon Valley widely assumes an AI-driven jobs apocalypse is inevitable, but researchers at Anthropic are examining actual data to test this narrative. One specific dataset could fundamentally reshape how we understand AI's real employment impact. This analysis matters for workforce planning and policy decisions.See more
14dUri Maoz
A
Allen Institute for Artificial Intelligence (AI2) - YouTubeWildDet3D - an open model for monocular 3D detection
WildDet3D is an open-source monocular 3D detection model released by AI2. Enables 3D object detection from single-image inputs. Technical resource for computer vision product teams.See more
13dAllen Institute for Artificial Intelligence (AI2) - YouTube
S
Socrates
Steven Kotler on We Are As Gods: Godlike Power, Stone Age Minds
14dSocrates
T
The Berkeley Artificial Intelligence Research Blog
Identifying Interactions at Scale for LLMs
1moThe Berkeley Artificial Intelligence Research Blog
B
Bala Priya CVector Databases Explained in 3 Levels of Difficulty
Vector databases store and query high-dimensional embeddings, enabling semantic search and similarity matching—a core capability for AI applications. The post explains the concept across three difficulty levels, from basic intuition to technical implementation details.See more
25dBala Priya C
L
Last Week in AILWiAI Podcast #238 - GPT 5.4 mini, OpenAI Pivot, Mamba 3, Attention Residuals
19dLast Week in AI