Google has introduced FACTS Grounding, a new benchmark system to evaluate how accurately large language models (LLMs) use source material in their responses. The benchmark comprises 1,719 examples across various domains including finance, technology, and medicine. The FACTS team at Google DeepMind and Google Research developed the system, which uses three frontier LLM judges – …
OpenAI has made its advanced o1 artificial intelligence model available to third-party developers through its API. According to an article by Carl Franzen in VentureBeat, this release represents a significant advancement in making sophisticated AI technology accessible to developers. The o1 model, first announced in September 2024, differs from traditional large language models by incorporating …
Nvidia and DataStax have introduced a new AI technology that reduces data storage requirements by 35 times for generative AI systems. As reported by Michael Nuñez for VentureBeat, the Nvidia NeMo Retriever microservices, integrated with DataStax’s AI platform, enables faster and more accurate information retrieval across multiple languages. The technology has already shown impressive results …
AI company Cohere has introduced Command R7B, a new compact language model designed for enterprise applications. According to VentureBeat reporter Taryn Plumb, the model supports 23 languages and specializes in retrieval-augmented generation (RAG). Command R7B outperforms similar-sized models from competitors like Google, Meta, and Mistral in mathematics and coding tasks. The model features a 128K …
Tokyo-based startup Sakana AI has created a breakthrough technique that reduces memory usage in large language models by up to 75%. As reported by Ben Dickson, the system called “universal transformer memory” uses neural attention memory modules (NAMMs) to efficiently manage information processing. These modules analyze the model’s attention layers to determine which information to …
Lambda, a San Francisco-based technology company, has introduced a new AI inference API service that promises the lowest costs in the industry. According to VentureBeat reporter Carl Franzen, the service allows enterprises to deploy AI models without managing computing infrastructure. The API supports various advanced models including Meta’s Llama 3.3 and Alibaba’s Qwen 2.5, with …
Microsoft has introduced a new AI model that delivers superior mathematical reasoning capabilities while using significantly less computing power than larger competitors. According to Michael Nuñez’s report in VentureBeat, the 14-billion-parameter Phi-4 model outperforms larger systems like Google’s Gemini Pro 1.5. The model excels particularly in mathematical problem-solving, achieving top scores on standardized math competition …
ServiceNow has launched Fast-LLM, an open-source framework that speeds up artificial intelligence model training by 20%. As reported by Sean Michael Kerner for VentureBeat, the technology has already proven successful in training ServiceNow’s StarCoder 2 language model. Fast-LLM introduces two key innovations: “Breadth-First Pipeline Parallelism” for optimized computation ordering and improved memory management that reduces …
Meta has released Llama 3.3 70B, a new large language model that achieves GPT-4 level performance while running on high-end consumer laptops. The breakthrough was documented by developer Simon Willison testing the model on a 64 GB MacBook Pro M2, demonstrating capabilities comparable to much larger models like Meta’s own Llama 3.1 405B. The model …
A comprehensive analysis reveals that artificial intelligence-assisted coding tools, while boosting developer productivity, are not necessarily leading to better software quality. Software engineer Addy Osmani writes that AI tools can help developers achieve about 70% of a project quickly but struggle with the crucial final 30% that makes software production-ready. The report identifies that experienced …