?? AI K-news #7
Here we go with a new edition of our newsletter with the most trend conversations around AI this week.
? Amazon and Databricks Partner to Slash AI Costs with Trainium Chips
Amazon and Databricks have signed a five-year agreement to utilize Amazon’s Trainium AI chips for developing AI systems, potentially cutting costs for businesses. The partnership enables Databricks to offer more affordable AI services by leveraging Amazon’s specialized chips, designed to be more cost-effective than Nvidia’s GPUs. This move strengthens Amazon's position in the AI market, helping companies like Edmunds and NinjaTech AI reduce expenses by up to 40% while building custom AI solutions.
More info: https://www.wsj.com/articles/amazon-databricks-strike-five-year-deal-around-ai-chips-a2a135ab
?? Mistral Advances Edge AI with Ministral 3B and 8B: Pushing Efficiency and Customization
Mistral AI has unveiled two new models, Ministral 3B and Ministral 8B, designed for edge computing and on-device applications. These models set a new standard in the sub-10B category, excelling in areas such as knowledge, commonsense reasoning, and function-calling. Both support up to 128k context length, with Ministral 8B featuring an interleaved sliding-window attention mechanism for faster and more efficient inference. Targeting privacy-first applications, from smart assistants to autonomous robotics, these models provide compute-efficient, low-latency solutions. Available today, Ministral models outperform previous versions and rival models across multiple benchmarks, ensuring real-time, low-cost AI performance.
More info: https://mistral.ai/news/ministraux/
?? OpenAI Unveils SWARM: The Lightweight Framework to Revolutionize Multi-Agent Systems
OpenAI has introduced SWARM, a cutting-edge, lightweight multi-agent orchestration framework designed to simplify multi-agent system development with OpenAI models. SWARM focuses on two main concepts: routines (agents) and handoffs (agent task transfers), offering a streamlined alternative to more complex frameworks like Hugging Face’s Transformers Agents 2.0. While this experimental and educational framework is open-source and licensed under MIT, it provides maximum transparency and control, empowering developers to customize and scale their projects effectively.
?? U.S. Government Cracks Down on AI Scams in “Operation AI Comply”
The U.S. Federal Trade Commission (FTC) has launched "Operation AI Comply," targeting businesses that allegedly misused AI technology to deceive consumers. Five companies are under scrutiny, including DoNotPay and Rytr, which settled charges of overstating their AI capabilities. Others, like Ascend Ecommerce and FBA Machine, face lawsuits for defrauding customers of millions. The crackdown aims to prevent exaggerated claims about AI’s potential and safeguard consumer trust in AI-driven products.
领英推荐
?? Harvard Develops Cutting-Edge AI Solution for Medical Queries
Harvard Medical School, in collaboration with Pfizer and other institutions, has introduced a groundbreaking methodology to enhance AI’s medical reasoning capabilities. Addressing the shortcomings of large language models (LLMs) in medical contexts, the new system integrates knowledge graphs with LLMs, improving accuracy in handling complex medical questions. This experimental, open-source framework aims to blend codified and non-codified medical knowledge, offering a new path for AI in healthcare.
More info: 2410.04660 ( arxiv.org )
? UNDP and IBM Launch AI Tools to Support Clean Energy Transition
The United Nations Development Programme (UNDP) and IBM have introduced two new AI-driven models to boost energy access and equity, now available on UNDP's GeoHub platform. The Electricity Access Forecasting and Clean Energy Equity Index models utilize IBM’s watsonx AI and cloud technologies to provide insights on electricity access forecasts and clean energy equity. These tools aim to empower policymakers and communities with data for informed, sustainable energy decisions, particularly in the Global South. The initiative is part of IBM’s Sustainability Accelerator program.
?? Liquid AI Unveils New Non-Transformer AI Model, Performs Well in Benchmarks
Liquid AI has introduced its Liquid Foundation models, a new family of generative AI models based on a non-Transformer architecture, aiming to challenge current AI benchmarks. Available in three sizes (1B, 3B, and 40B parameters), the models show strong performance in memory efficiency and benchmarks like MMLU Pro. However, real-world testing revealed mixed results, particularly in logic and reasoning tasks, leaving room for improvement in practical applications despite promising initial benchmarks.
?? Nvidia's Nemotron AI Model Surpasses GPT-4 in Benchmarks
Nvidia has launched its new open-source AI model, Llama-3.1-Nemotron-70B-Instruct, claiming it outperforms advanced AI models like GPT-4o and Claude-3. Built on Meta's Llama-3.1-70B model, Nvidia enhanced it using proprietary datasets, fine-tuning methods, and cutting-edge AI hardware. According to Nvidia, the model excels in chatbot performance tests, scoring highly in the "Hard" automated benchmark, making it one of the most "useful" models for tasks like coding and problem-solving.
This edition was edited by Jesus G. and Héctor Castillejo