# Mar 20, 2025

[492 days ago](https://news.lambda.ai/news/2025-03-21 "Navigate to 492 days ago")

## Daily

## Weekly

- **Orpheus-3B** – Emotive TTS by Canopy Labs
  - **Canopy Labs** offers advanced AI solutions that enhance data-driven decision-making for businesses, focusing on user engagement and predictive analytics.

- **Bolt3D**: Generating 3D Scenes in Seconds
  - **Bolt3D** achieves **feed-forward 3D scene generation** in just **6.25 seconds** using a single GPU, showcasing its efficiency in creating complex visuals rapidly.

- **SoftBank Group** to Acquire **Ampere Computing** for **6.5B**
  - **SoftBank Group** will acquire **Ampere Computing** for **$6.5 billion**, enhancing its AI infrastructure and semiconductor capabilities, crucial for advancing **Artificial Super Intelligence**.

- **Hunyuan3D-2-Turbo**: fast high-quality shape generation in ~1s on a 4090
  - **Release of new models**: Tencent has launched **Hunyuan3D-2-Turbo**, **Hunyuan3D-2mini-Turbo**, and **FlashVDM**, enhancing capabilities in 3D generation and processing, with links to their respective models available on Hugging Face [Hunyuan3D-2-Turbo](https://huggingface.co/tencent/Hunyuan3D-2/tree/main/hunyuan3d-dit-v2-0-turbo), [Hunyuan3D-2mini-Turbo](https://huggingface.co/tencent/Hunyuan3D-2mini/tree/main/hunyuan3d-dit-v2-mini-turbo), and [FlashVDM](https://github.com/Tencent/FlashVDM).

- **DESI Opens Access to the Largest 3D Map of the Universe Yet**
  - The **Dark Energy Spectroscopic Instrument (DESI)** has released its first 13 months of data, comprising **270 terabytes** that detail distances to **18.7 million celestial objects**, including **4 million stars** and **13.1 million galaxies**.

- **ChatGPT** hit with **privacy complaint** over **defamatory hallucinations**
  - **OpenAI faces a serious privacy complaint** in Europe due to ChatGPT's **hallucination of defamatory information**, including false claims of murder against an individual, raising concerns about the accuracy of AI-generated content.

- **Google calls Gemma 3** the **most powerful AI model** you can run on one GPU
  - **Gemma 3** is touted as the **most powerful AI model** operable on a single GPU, capable of interpreting **text, images, and short videos** across over 35 languages, enhancing its utility for developers.

- **Want even tinier chips? Use a particle accelerator**
  - **Particle accelerators** may revolutionize chip manufacturing by enabling the etching of **nano-scale designs** with high-speed electrons, potentially surpassing current optical lithography methods.

- **RWKV-7 "Goose"** with **Expressive Dynamic State Evolution**
  - **RWKV-7 "Goose"** achieves **state-of-the-art performance** in multilingual tasks at the **3 billion parameter scale**, outperforming other models trained on more data while maintaining constant memory and inference time per token.

- **Claimify**: Extracting high-quality claims from language model outputs
  - **Claimify** is a novel LLM-based method that enhances **claim extraction** from language model outputs, addressing common inaccuracies and omissions found in previous systems.

- **Analyzing Failure Modes in Sliding Window-Based Time Series Clustering**
  - The paper reveals that **small sliding windows yield flat centroids**, while **large windows** effectively form **interval clusters**, providing a mathematical basis for observed clustering behaviors in time series data.

- **Evaluating Video Models on Impossible Scenarios**: A Benchmark for Generation and Understanding of Counterfactual Videos
  - **IPV-Bench** is a new benchmark assessing video generation models' grasp of **physics** through 1,000 prompts that challenge their handling of **physically impossible scenarios** across nine categories, revealing significant gaps in understanding.

- **How 'inference' is driving competition to Nvidia's AI chip dominance**
  - **Nvidia's dominance in AI chips is challenged** as competitors like DeepSeek and others focus on **inference**, which is becoming a critical demand in AI applications, shifting from traditional training models.

- **GTC 2025** – Announcements and Live Updates
  - **GTC 2025** will showcase cutting-edge advancements in **AI, robotics, and accelerated computing**, featuring influential speakers like **Yann LeCun** and **Frances Arnold**, who will challenge conventional thinking and inspire innovation.

- **NVIDIA Blackwell Powers Real-Time AI for Entertainment Workflows**
  - **NVIDIA's RTX PRO Blackwell GPU series** revolutionizes real-time AI in media workflows, enhancing content creation and live video processing with **up to 4,000 AI trillion operations per second** and **96GB of GDDR7 memory** for handling complex datasets.