# Jun 5, 2025

- **Cloud Run GPUs are now generally available**, enabling developers to run AI workloads with **NVIDIA L4 GPUs** without needing quota requests, thus simplifying access to GPU acceleration for serverless applications.

- **TU Delft's autonomous drone** triumphed at the A2RL Drone Championship, marking a historic first by defeating human champions with speeds reaching **95.8 km/h** on a challenging track, showcasing advanced AI capabilities in real-world conditions.

- **Gemini 2.5 Pro** is Google's most advanced AI model, designed for **coding** and **complex prompts**, featuring enhanced reasoning capabilities through its new **Deep Think** mode, which employs cutting-edge parallel thinking techniques.

- **AGI cannot be achieved through multimodal models** as they fail to address the essential need for a physical understanding of the world, which is crucial for solving real-world problems like motion planning and social coordination.

- **VLMs struggle** with temporal patterns when spatial information is obscured, as demonstrated by the introduction of **SpookyBench**, where humans achieve over **98% accuracy** while VLMs score **0%**.

- **LLMs exhibit a strong bias towards aggressive statistical compression**, which contrasts with human systems that prioritize _adaptive nuance_ and _contextual richness_, even at the cost of compressional efficiency.

- **Eleven v3 (alpha)** introduces a **highly expressive Text to Speech model** that allows for dynamic conversations and emotional depth through audio tags, enhancing user engagement and realism in generated speech.

- **Large language models (LLMs) exhibit significant variability in their ability to accurately cite and comprehend sources, with some models** demonstrating a tendency for "link hallucination," where they generate incorrect or fabricated citations. This discrepancy raises concerns about the reliability of LLMs in critical applications such as fact-checking and information retrieval.

- **NVIDIA Blackwell architecture** achieved **breakthrough performance** in the latest MLPerf Training, outperforming previous generations by **2.2x** on the Llama 3.1 405B pretraining benchmark, showcasing its capability to handle demanding AI workloads.

- **Chain-of-Zoom (CoZ)** revolutionizes single-image super-resolution (SISR) by enabling extreme magnification beyond 256x through an autoregressive framework that utilizes multi-scale-aware prompts, enhancing visual fidelity without additional training.

- **Microsoft's Aurora model** can deliver accurate **10-day weather forecasts** rapidly, showcasing its potential to transform not just weather predictions but also other Earth systems like air pollution and wave height.

- **Nvidia's Blackwell GPUs** have achieved unprecedented success in the latest MLPerf training results, outperforming competitors across all six benchmarks, while AMD's MI325X has shown competitive performance on the popular LLM fine-tuning benchmark.

- **Neuromorphic computing** aims to create AI systems that operate on just **20 watts**, mimicking the energy efficiency of the human brain, potentially revolutionizing AI scalability and adaptability.

- The **Targeted Information Forgetting (TIF)** framework introduces a dual approach to unlearning in Large Language Models (LLMs), effectively distinguishing between **unwanted words (UW)** and **general words (GW)** to enhance model utility while mitigating over-forgetting.

- **ATLAS** introduces a **long-term memory module** that optimally memorizes context by leveraging both current and past tokens, significantly enhancing performance in autoregressive language modeling tasks.