AI in practice Archive

Jul 11, 2025

Mistral AI and All Hands AI have introduced two new models designed for AI-powered programming agents: Devstral Small 1.1 and Devstral Medium. Devstral Small 1.1 2507 is open source and can run locally on an RTX 4090 or a Mac with 32 GB of RAM. It achieved a 53.6% score on the SWE-Bench Verified benchmark and supports XML along with other formats.

Devstral Medium scored 61.6% on the same benchmark. According to Mistral, it offers more power and a lower price than Gemini 2.5 Pro and GPT-4.1. The model is available via API, supports fine-tuning, and will soon be integrated into Mistral Code.

Comment

Matthias Bastian

Jul 10, 2025

AI in practice

Google is adding three new AI modes to Firebase Studio. The update introduces "Agent" modes powered by Gemini 2.5: a conversational "Ask" mode, a guided "Agent" mode that acts with user approval, and a fully autonomous "Agent Auto-run" mode that can write and update code on its own. Google says all three modes respect project-specific rules and require user approval for any security-sensitive actions.

Google says that in Agent Auto-run mode, Firebase Studio can generate entire apps or add new features to existing projects with minimal user input. | Video: Google

Firebase Studio now also supports the Model Context Protocol (MCP), making it easier to connect external data sources. Developers can access the Gemini command line (CLI) directly in the terminal for tasks like debugging and code management.

Comment

Matthias Bastian

Jul 10, 2025

AI in practice

Amazon has secured multi-year licensing deals with Condé Nast and Hearst. Content from magazines like Good Housekeeping, Cosmopolitan, and Harper's Bazaar will be added to its AI shopping assistant Rufus, Digiday reports. The move expands Rufus's offerings with trusted editorial content to help shoppers make informed decisions. Condé Nast confirmed the new partnership, following a similar agreement Amazon reached with The New York Times in May, which includes material from NYT Cooking and The Athletic. Financial terms were not disclosed. The first content from these publishers is set to appear in Rufus this summer. Rufus uses Amazon's product catalog and web data to help shoppers make purchasing decisions with AI.

Comment

Maximilian Schreiner

Jul 10, 2025

AI in practice

Google has added a new feature to Veo 3, its AI video app, that lets users turn a single image into a short video. Anyone with a Pro or Ultra subscription can upload a photo in the Gemini app or on gemini.google, choose "Video" in the prompt bar, and describe what should happen in the clip. According to Google, the rollout starts today, though availability may vary - users can check directly in the app to see if they have access.

✨?✨ Special delivery! A new Gemini feature just dropped. Make photos come alive by turning them into videos with sound.

- Google Gemini App (@GeminiApp) July 10, 2025

Comment

Maximilian Schreiner

Jul 10, 2025

AI in practice

Musk unveils Grok 4 as xAI’s new AI model that beats OpenAI and Google on major benchmarks

While X has become a breeding ground for fake news and hate speech under Elon Musk's leadership, his own AI chatbot seems to be fighting back.

Matthias Bastian

Jul 9, 2025

Update

AI in practice

Elon Musk blames "user prompts" for Grok's wave of anti-Semitic posts and Hitler praise

Matthias Bastian

Jul 9, 2025

AI in practice

Google is rolling out Gemini to smartwatches running Wear OS 4 or later, including devices from Pixel, Samsung, OPPO, OnePlus, and Xiaomi. Users can launch the AI assistant with a voice command, side button, or app icon to get answers right on their wrist - from checking the weather and calendar appointments to reading emails. If permissions are granted, Gemini pulls information from Google services like Gmail, Calendar, and Maps. Voice controls also let users manage reminders and messages.

The feature will be available globally over the next few weeks, with partial support for manufacturer-specific apps coming later with Wear OS 6.

Comment

Matthias Bastian

Jul 9, 2025

AI in practice

Anthropic is rolling out new integrations for its Claude AI chatbot, connecting it with the Canvas learning platform, Panopto, and Wiley. Students can access lecture recordings, academic literature, and other course materials directly in Claude. Wiley supplies scientific resources, while Panopto provides lecture transcripts. The integration uses Anthropic's own Model Context Protocol (MCP), and the connection with Canvas is based on the Learning Tools Interoperability (LTI) standard.

Anthropic says all conversations are private by default and aren't used for training. Partners include the University of San Francisco School of Law and Northumbria University.

Comment

Matthias Bastian

Jul 9, 2025

AI in practice

Perplexity has introduced "Comet," an AI-powered browser designed to replace traditional tabs with an interface that uses LLMs to help users navigate the web. With Comet, users can write emails, plan meetings, compare products, ask questions, or highlight text to get instant explanations. Perplexity says the goal is to make browsing simpler and give people better access to information. Comet will launch first for Perplexity Max subscribers at $200 per month, with access managed through a waitlist and invitation system starting this summer.

Video: Perplexity

Comment

Maximilian Schreiner

Jul 9, 2025

AI in practice

Replit is teaming up with Microsoft to automate app development for non-programmers.

The company is positioning itself as a no-code prototyping platform for business users, promising to turn natural language prompts into working apps - complete with database, authentication, and storage. These applications are available directly through the Azure Marketplace and run on Azure infrastructure managed by Replit.

The partnership is aimed squarely at the prototyping and design market, an area currently dominated by Figma. Replit does not see itself as direct competition for Microsoft's flagship AI tool, GitHub Copilot.

Comment