Google releases Gemma 4 open AI model for complex logic and agentic workflows.

·1 views

Google has officially released Gemma 4, a new open-model AI family. The model is designed for complex reasoning, agentic workflows, and offers multimodal capabilities with a 256K context window. It is currently available for deployment on Google Cloud, Vertex AI, and other platforms. This announcement is separate from industry reports about a future flagship model called Gemini 4, which remains unconfirmed by Google.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer tests in-browser translation, finds 30% of phrases wrong in meaning

A developer tested a free, privacy-focused browser-based machine translation tool that processes text locally. Using 30 English-Russian phrases, the evaluation found 30% had a meaning-altering error, while 43% were fully correct. The tool struggled significantly with idioms, casual speech, and ambiguous legal or technical phrases. Following benchmarks against several API-based models, the developer replaced the engine with a server-side model for improved accuracy, though this sacrifices the original privacy benefit.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer creates free 563-drug harm reduction library using Cloudflare's free tier

A developer has created plugreports, a free harm-reduction library containing profiles for 563 drugs and 61 mixing danger guides in nine languages. The platform provides verified hotlines across nine regions and an identification video library to offer factual information instead of forum speculation. The entire system runs on Cloudflare's free tier using a custom Python script that generates approximately 1,100 static HTML pages with structured data. Machine translation via Workers AI handled bulk translation of content, with human review possible through an admin panel. The project was motivated by search data showing high volume for drug identification and safety questions where inaccurate information could be dangerous.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

OpenAI launches Dots agents, developer creates open-source local alternative called Sparks

OpenAI launched Dots, persistent background AI agents, on Tuesday. A developer created Flash, an open-source alternative that runs locally within 24 hours. The developer's version, called Sparks, operates on user-selected models through Ollama and continues working after the application closes. Sparks are configured with specific goals, schedules, and boundaries via simple commands or requests. Each spark functions autonomously, files reports, and can be interacted with through a chat interface.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI governance must authorize actions before execution, not just audit after

Current AI governance tools mainly audit what a model produced after the fact. This is insufficient when AI agents perform consequential actions like moving money or changing records. Effective control requires authorizing each specific action before the external system state changes. The decision must validate current authority, policy, evidence, and dependencies immediately before execution. Governance systems must reliably produce clear PERMIT, BLOCK, or ESCALATE decisions based on fresh, verified facts.

0 comments Read more at DEV Community