Foundation models in pathology are failing not due to size or training duration but because they are built on flawed assumptions about data scalability and generalization. Clinical performance has plateaued, as models struggle with variability across institutions and real-world applications, highlighting a need for task-specific approaches instead of generalized solutions. Alternative methods, like weakly supervised learning, have shown promise in achieving high accuracy without the limitations of foundation models.

Text-to-LoRA (T2L) is a hypernetwork that enables the instant adaptation of large language models to specific tasks using only natural language descriptions, eliminating the need for extensive fine-tuning and dataset curation. Trained on various pre-existing LoRA adapters, T2L can generate task-specific adapters in a single forward pass, demonstrating performance comparable to traditional methods while significantly reducing computational requirements and allowing zero-shot generalization to new tasks.

Apple has introduced the Foundation Models Framework at WWDC, enabling developers to integrate powerful on-device AI capabilities into their apps. This framework emphasizes privacy and offline functionality, allowing applications like education and outdoor apps to utilize Apple's AI without incurring inference costs. Developers can start testing the framework today through the Apple Developer Program, with a public beta expected next month.

Mark Zuckerberg announced the establishment of Meta Superintelligence Labs (MSL), which will focus on advancing AI superintelligence and be led by key hires including Alexandr Wang and Nat Friedman. The initiative comes as part of a significant investment into AI talent, aiming to enhance Meta's capabilities in developing advanced foundation models and products.

ChatGPT is nearing 700 million weekly active users, prompting developers to focus on building solutions that complement rather than compete with dominant AI platforms. Emphasizing the importance of targeting niche applications—what the big players cannot or will not address—can lead to market success, as illustrated by examples like Harvey and Writer. The key is to identify and own the "side quests" that provide real value in specialized areas.

Amazon has introduced Amazon Nova, a new generation of foundation models that offer advanced intelligence and competitive pricing. The company is expanding its Artificial General Intelligence (AGI) efforts with a new lab in San Francisco, seeking diverse talent to contribute to innovative AI solutions that address real-world challenges.

The article discusses the emerging role of foundation models in processing tabular data, highlighting their potential to improve data analysis and machine learning tasks. It examines the benefits of leveraging these models to enhance predictive performance and streamline workflows in various applications. Additionally, the article explores the challenges and future directions for integrating foundation models in tabular datasets.

Apple's Foundation Models framework, introduced with iOS 26, empowers developers to create innovative, privacy-focused AI features for apps, enabling offline functionality and cost-free AI inference. Apps across health, education, and productivity are harnessing this technology to enhance user experiences, personalize interactions, and improve data management, all while ensuring user privacy.

Stanford's Marin project aims to redefine openness in AI by providing complete transparency throughout the foundation model development process, including sharing code, datasets, and training methodologies. Utilizing JAX and a new framework called Levanter, the project addresses key engineering challenges to ensure reproducibility and efficiency in training large-scale models. By fostering collaboration and trust, the Marin project invites researchers to participate in advancing foundation model research.