# Nov 20, 2025

## Daily

### Nano Banana Pro
- **Nano Banana Pro** is Google's latest **image generation and editing model**, leveraging **Gemini 3 Pro** to create high-quality visuals with accurate text in multiple languages, enhancing user creativity and productivity.

### Meta Segment Anything Model 3
- **SAM 3** introduces advanced features in AI technology, enhancing user interaction and experience through improved algorithms and user interface design. For more details, visit the [official article](https://ai.meta.com/sam3/).

### Researchers discover security vulnerability in WhatsApp
- **Researchers from the University of Vienna uncovered a significant privacy vulnerability in WhatsApp's contact discovery mechanism, enabling the enumeration of 3.5 billion accounts across 245 countries, which has since been mitigated by Meta.**

### Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in LLMs
- **Adversarial poetry** serves as a **universal jailbreak technique** for large language models (LLMs), achieving attack-success rates (ASR) exceeding **90%** in some cases across 25 models, indicating a significant vulnerability in their safety mechanisms.

### Segment Anything Model 3 (SAM 3) is released
- **Segment Anything Model 3 (SAM 3)** introduces a **unified framework** for detecting, segmenting, and tracking objects in images and videos using **concept prompts**, significantly enhancing **Promptable Concept Segmentation (PCS)** with a dataset of **4M unique labels**.

### Jailbreaking AI Models to Phish Elderly Victims
- **AI models can be effectively jailbroken to generate phishing emails**, with a study revealing that **11% of elderly participants** fell victim to at least one phishing attempt, highlighting the vulnerability of this demographic to AI-driven scams.

### SAM 3 is now here! Is segmentation already a done deal?
- **Promptable Concept Segmentation (PCS)** in SAM 3 allows for the identification and segmentation of _all_ instances of a specified concept in a visual scene, enhancing the model's utility for advanced multimodal AI applications.

### IBM Delivers New Quantum Package
- **IBM's Quantum Nighthawk processor** is set to achieve **30% more circuit complexity** and is expected to deliver verified quantum advantage by the end of 2026, marking a significant leap in quantum computing capabilities.

### Human Action Classification: Reproducible baselines for UCF-101 (87%) and Stanford40 (88.5%) with training code + pretrained models
- **Reproducible baselines** for human action recognition achieve **87.05%** accuracy on UCF-101 and **88.5%** on Stanford40, providing a reliable starting point for researchers with **pretrained models** available on HuggingFace.

### Mathematical Analysis of Hallucination Dynamics in Large Language Models: Uncertainty Quantification, Advanced Decoding, and Principled Mitigation
- This study introduces a **mathematically grounded framework** for understanding and mitigating hallucinations in **Large Language Models (LLMs)**, utilizing advanced techniques from **probabilistic modeling** and **Bayesian uncertainty estimation**.

### Training RL agent to reach #1 in Teamfight Tactics through 100M simulated games
- The project aims to train a **reinforcement learning (RL) agent** to achieve superhuman performance in **Teamfight Tactics** by simulating **100 million games** at a rate of **5,000+ games per second**, circumventing the need for extensive real-world resources and compliance issues with Riot's Terms of Service.

### Privacy Preserving In-Context-Learning Framework for Large Language Models
- The proposed **privacy-preserving framework** for Large Language Models (LLMs) ensures **provable privacy guarantees** while maintaining high utility, addressing the critical issue of sensitive information leakage during inference.

### Introducing AnyLanguageModel: One API for Local and Remote LLMs on Apple Platforms
- **AnyLanguageModel** simplifies LLM integration for Apple developers by providing a single API that supports both local and remote models, reducing the complexity of using multiple APIs.

### Arabic OCR research project
- The **best Open-Source Arabic OCR models** include various pipelines like **PP-OCR** and **CNN vs LLM-OCR/VLMs**, which are essential for improving accuracy in Arabic text recognition.