Stanford University researchers have developed a new framework called Item Response Scaling Laws (IRSL), which significantly reduces the computational demand for training large language models. By using statistical concepts from measurement science and education, IRSL can cut computational demand by over 99% while maintaining or improving predictive accuracy. This approach could make AI training more accessible to researchers and smaller developers.
Cohere has released Command A+, a 218B sparse MoE model optimized for agentic workflows, reasoning, and multimodal document processing. This model achieves high performance with minimal compute overhead and can run on as few as two H100 GPUs. Meanwhile, Meta has been using its employees as AI training data, raising questions about the ethics of using employee data for AI development.
Several companies are introducing new AI-powered tools and services. Traini has launched its Pet Emotional Intelligence API, which interprets animal behavior and emotions through multimodal AI. Airbnb is expanding its services to include airport pickups, luggage storage, car rentals, and AI-powered travel tools. UneeQ has introduced AI Buddy, a digital human that provides face-to-face, spoken coaching to sales teams.
The next AI boom is expected to be in worker training, as companies are not doing enough to prepare employees for AI-powered workplaces. A report found that 48% of US workers in educated professional roles are not prepared to use AI tools effectively. Employers need to invest in training programs to help workers adapt to AI-driven changes.
Key Takeaways
• Stanford researchers developed IRSL, a framework that reduces computational demand for large language model training by over 99%.\n• Cohere released Command A+, a 218B sparse MoE model for agentic workflows and multimodal document processing.\n• Meta used its employees as AI training data, then laid them off.\n• Traini launched its Pet Emotional Intelligence API for interpreting animal behavior and emotions.\n• Airbnb expanded services to include airport pickups, luggage storage, car rentals, and AI-powered travel tools.\n• UneeQ introduced AI Buddy, a digital human for sales coaching.\n• The next AI boom is expected to be in worker training, with 48% of US workers not prepared to use AI tools effectively.\n• Ace Hardware deployed an AI-powered assistant called 'Hey ARMA' to help store associates serve customers.\n• The Data Rules Project aims to improve data governance and change rules governing data access, use, and sharing across borders.Stanford Researchers Develop IRSL for Cheaper AI Training
Stanford University researchers have introduced IRSL, a framework that reduces computational demand in large language model development. IRSL draws on measurement science and educational testing methods, similar to standardized exams. This approach can reduce computational demand by over 99% while maintaining or improving predictive accuracy. The method uses carefully selected questions to estimate model capabilities more efficiently. This could make AI training cheaper and more accessible to researchers and smaller developers.
New AI Approach Changes How Models Are Trained
Researchers at Stanford have developed Item Response Scaling Laws (IRSL), a new approach to scaling laws that reduces training demands significantly. IRSL uses statistical concepts from measurement science and education to tailor scaling algorithms. This approach achieves equal or greater predictive accuracy with far fewer queries, saving time and money. The framework is inspired by standardized academic assessments like the SAT.
Traini Launches Pet Emotional Intelligence API
Traini has introduced its Pet Emotional Intelligence API, which interprets animal behavior and emotions through multimodal AI. The platform uses data from over 2 million dogs and can be embedded into third-party products and services. Traini also unveiled Sentra, an AI-enabled cognitive smart collar that provides real-time emotional insights. The company has secured a strategic manufacturing partnership and has seen strong commercial interest in its API.
Airbnb Expands Services with Airport Pickups and AI Tools
Airbnb is expanding its services to include airport pickups, luggage storage, car rentals, and AI-powered travel tools. The company has added boutique hotels and exclusive travel experiences, including tours tied to landmarks. Airbnb's new features aim to enhance the travel experience and provide more services to its users.
The Next AI Boom Is Worker Training
The next AI boom is expected to be in worker training, as companies are not doing enough to prepare employees for AI-powered workplaces. A report found that 48% of US workers in educated professional roles are not prepared to use AI tools effectively. Employers need to invest in training programs to help workers adapt to AI-driven changes.
Data Rules Project Aims to Improve Data Governance
The Data Rules Project examines how global data governance shapes the development and deployment of AI in pharmaceutical research and biotechnology. The project aims to change the rules governing data access, use, and sharing across borders.
Cohere Releases Command A+ for Agentic Workflows
Cohere has released Command A+, a 218B sparse MoE model for agentic workflows. The model is optimized for reasoning, agentic workflows, and multimodal document processing. It achieves high performance with minimal compute overhead and can run on as few as two H100 GPUs.
UneeQ Unveils AI Buddy for Sales Coaching
UneeQ has introduced AI Buddy, a digital human that provides face-to-face, spoken coaching to sales teams. AI Buddy uses 3D analysis and offers conversational feedback in real-time, helping sales teams improve their performance.
Ace Hardware Introduces AI-Powered Assistant
Ace Hardware has deployed an AI-powered assistant called 'Hey ARMA' to help store associates better serve customers. The tool provides product knowledge, project guidance, and recommendations, allowing associates to spend more time engaging with customers.
Meta Used Employees for AI Training
Meta used its employees as AI training data, then laid them off. The company used an unusual training strategy, learning by watching employees and replacing them afterward.
Sources
- New Stanford scaling method could make AI training cheaper
- New Approach to Scaling Laws Could Change How AI Models Are Trained
- Traini Launches Pet Emotional Intelligence API, Expands AI Hardware and Partnerships
- Airbnb launches major expansion with airport pickups, luggage storage and AI-powered travel tools
- The Next AI Boom Is Worker Training
- The Data Rules Project
- Cohere Releases Command A+: A 218B Sparse MoE Model for Agentic Workflows That Runs on as Few as Two H100 GPUs
- UneeQ Unveils AI Buddy at the Gartner CSO & Sales Leader Conference 2026
- Ace Hardware debuts AI-powered assistant
- 😺 Meta used staff as AI training data. Then cut them.
Comments
Please log in to post a comment.