Google introduces TorchTPU, a solution enabling native PyTorch on TPUs, allowing developers to migrate their ML workloads to Google’s supercomputing infrastructure with minimal code changes and full utilization of TPU resources.
Google demonstrates how to build long-lived AI agents with the Agent Development Kit that work for weeks, pause, and retain their memory – without the context loss of traditional chatbots.
Google’s new Gemini Embedding 2 unifies text, images, video, audio and documents in a single embedding model, supporting over 100 languages and processing multiple media types simultaneously to enable new multimodal AI applications and intelligent search systems.
Google Genkit introduces a new middleware system for extending and securing AI applications, with modular hooks enabling retries, fallbacks, and human oversight—available in TypeScript, Go, and Dart, with Python support coming soon.
Google extends MaxText with Supervised Fine-Tuning and Reinforcement Learning for single-host TPUs, enabling efficient post-training of language models on v5p-8 and v6e-143 systems.
Google enhances the Google Pay API with new features for merchant-initiated payments, enabling better control over subscriptions, deferred payments and auto-reloads, plus increased transparency for users.
Google experts show in their AI Agent Clinic how fragile AI agents are made production-ready — from cost control through error handling to scaling for real-world requirements.
ADK’s SkillToolset enables AI agents to dynamically load domain-specific knowledge at runtime; progressive disclosure saves tokens and integrates context information strategically. A developer guide presents four practical application patterns.
Google launches Gemini Embedding 2, the first multimodal embedding model that connects text, images, videos, audio and documents in a unified space, supporting over 100 languages and enabling agent-based RAG applications and visual search.
Starting March 2024, Google is allowing individual users to change their account username. The old email remains as an alternative and uses the same inbox. App developers should check whether their systems support this change.
Google develops new method for LLM acceleration on TPUs using a diffusion-inspired approach — threefold speedup through parallel token prediction instead of sequential bottleneck.
Google Pay API receives new features for merchant-initiated transactions, with updates enabling explicit specification of payment conditions for subscriptions and deferred payments while providing users greater transparency.
