Google AI Releases MedGemma-1.5 | Anthropic Releases Cowork

Here is your today’s AI Dev Brief from Marktechpost, covering core research, models, infrastructure tools, and applied updates for AI developers and researchers.

Google has released MedGemma 1.5 and MedASR as open components in its Health AI Developer Foundations program, giving developers a practical starting point for medical imaging, text and speech workflows. MedGemma-1.5-4B is a multimodal model that supports text, two dimensional images, three dimensional CT and MRI volumes and whole slide pathology, with accuracy gains for disease findings and histopathology that reach or match strong task specific baselines. It also improves MedQA and EHRQA scores, which makes it suitable as a backbone for clinical question answering and chart summarization pipelines. MedASR is a Conformer based medical speech recognition model that reduces word error rate by 58 percent on chest X ray dictation and 82 percent on broader medical dictation benchmarks compared to Whisper large v3, providing a domain tuned speech front end for MedGemma centered applications....... Read the full analysis/article here.

Google AI released the Universal Commerce Protocol as an open standard that lets agents move from product search to secure checkout inside a single conversation, by giving platforms, merchants, payment services, and credential providers a shared capability based schema for discovery, checkout, and order management. UCP replaces bespoke retail integrations with a manifest based model, where agents discover merchant capabilities from a well known profile and negotiate supported extensions such as discounts or fulfillment, then invoke them over REST, Model Context Protocol, or Agent to Agent transports. Payments plug in through Agent Payments Protocol so each transaction is backed by cryptographic proof of user consent while merchants remain the Merchant of Record. This turns commerce into a predictable protocol surface so they can focus on ranking, policy, and user experience rather than rebuilding checkout logic for every retailer...... Read the full analysis/article here.

Anthropic has introduced Cowork, a research preview in the Claude macOS app that turns Claude into a local folder scoped agent for non coding work. Cowork is available to Claude Max subscribers and lets you grant Claude access to a chosen folder so it can read, edit, and create files, for example organizing downloads, building expense spreadsheets from screenshots, or drafting reports from scattered notes. Cowork runs on the same foundations as Claude Code and the Claude Agent SDK, but exposes them through a GUI with connectors, skills for documents and presentations, and optional pairing with Claude in Chrome for browser steps. It plans and executes multi step tasks with higher autonomy than regular chat, queues work in parallel, and asks before significant actions..... Read the full analysis/article here.

▶ [Open Source] Rogue: An Open-Source AI Agent Evaluator worth trying Codes & Examples

▶ How to Build a Multi-Turn Crescendo Red-Teaming Pipeline to Evaluate and Stress-Test LLM Safety Using Garak Codes Tutorial

▶ A Coding Implementation for an Agentic AI Framework that Performs Literature Analysis, Hypothesis Generation, Experimental Planning, Simulation, and Scientific Reporting Codes Tutorial

▶ How to Build a Neuro-Symbolic Hybrid Agent that Combines Logical Planning with Neural Perception for Robust Autonomous Decision-Making Codes Tutorial

▶ How to Design a Mini Reinforcement Learning Environment-Acting Agent with Intelligent Local Feedback, Adaptive Decision-Making, and Multi-Agent Coordination Codes Tutorial

100+ more open codes/notebooks here ➡️