A warm welcome to our new subscribers! Twice a week, Data Points brings you the latest AI news, tools, models, and research in brief. In today’s edition, you’ll find:
- Anthropic makes batch processing cheaper
- A new leaderboard measures models’ finance performance
- Aria introduces an new open multimodal MoE model
- SETI’s new AI model helps search for life in outer space
But first:
Open-source video generator uses novel techniques to produce high-quality clips
Pyramid Flow, a new open-source AI video generation model, launched this week, offering video clips up to 10 seconds long at 1366x768 resolution. Developed by researchers from Peking University, Beijing University of Posts and Telecommunications, and Kuaishou Technology, the model uses a novel technique called pyramidal flow matching to efficiently generate videos from text or image prompts. This model could help push proprietary AI video generators, providing developers and AI filmmakers with a free, open-source alternative for advanced video generation capabilities. (GitHub)
Gradio 5 launches with major upgrades for ML web app developers
Gradio 5 introduces server-side rendering for faster loading, refreshed design elements, low-latency streaming capabilities, and an experimental AI Playground for generating and modifying Gradio apps. The update addresses common developer concerns about performance, aesthetics, real-time functionality, and AI integration while maintaining a simple API and improving web security. This release positions Gradio as a production-ready framework for machine learning applications, with plans for future changes including multi-page apps, mobile support, and expanded component options. (Hugging Face)
Anthropic launches batch API for bulk processing
Anthropic introduced a new Message Batches API that allows developers to process up to 10,000 queries asynchronously at a 50 percent discount compared to standard API calls. The API supports Claude 3.5 Sonnet, Claude 3 Opus, and Claude 3 Haiku models, with batches processed within 24 hours, offering enhanced throughput and scalability for large-scale data processing tasks. This new feature aims to make completions more cost-effective for AI developers working with large datasets or complex analytical projects. (Anthropic)
Finance-focused leaderboard reveals surprising model performance
A new Open FinLLM Leaderboard evaluates language models specifically for financial tasks like stock prediction and credit risk assessment. The leaderboard uses zero-shot evaluation on real-world financial datasets to test models’ ability to generalize to unseen tasks without fine-tuning. GPT-4 and Llama 3.1 lead in many tasks, but smaller models like Llama-3.1-7b surprisingly outperformed larger counterparts in stock movement predictions, highlighting the importance of task-specific evaluations over model size. (Hugging Face)
Aria’s MoE model offers competitive performance while activating fewer parameters
Rhymes AI released Aria, an open-source multimodal Mixture-of-Experts model with impressive performance across various tasks. Aria activates only 3.9 billion of its 24.9 billion parameters per token, making it more efficient than comparable models while remaining competitive with proprietary systems like GPT-4 and Gemini 1.5. Aria's open-source nature and capabilities across language, vision, and coding tasks make it an intriguing choice for developers looking for a smaller, resource-efficient, and open multimodal model. (Rhymes AI)
SETI uses new AI processing models to detect and analyze radio signals
Scientists at the SETI Institute applied AI to detect faint radio signals from space in real time, building their own model using NVIDIA’s Holoscan platform and IGX Orin edge computing system. The team successfully captured and analyzed nearly 100Gbps of data from 28 antennas pointed at the Crab Nebula, doubling their previous processing speed. This use of AI in radio astronomy opens up new possibilities for analyzing streaming astronomical data and could transform how telescopes are used with AI software for space exploration. (SETI)
Still want to know more about what matters in AI right now?
Read this week’s issue of The Batch for in-depth analysis of news and research.
This week, Andrew Ng celebrated the 2024 Nobel Prizes in Physics and Chemistry being awarded to pioneers in AI, recognizing the significant contributions of Geoff Hinton, John Hopfield, Demis Hassabis, John Jumper, and David Baker. He expressed excitement about the growing recognition of AI’s impact on various fields and reflected on the importance of celebrating innovators within the AI community.
“It’s remarkable that the Nobel committees for physics and chemistry, which are made up of scientists in those fields, chose to honor AI researchers with this year’s awards. This is a sign of our field’s growing impact on society.”
Read Andrew’s full letter here.
Other top AI news and research stories we covered in depth: Meta debuts Movie Gen for text-to-video generation; OpenAI unveils tools for speech, vision, and cost-efficiency for GPT-4o API at DevDay; a German court rules that LAION did not violate copyrights, marking a win for AI in legal disputes; and researchers expose a black market for AI-driven cybercrime services.